SciELO - Scientific Electronic Library Online

 
vol.27 issue1Power Spectral Analysis of Bioacoustic Signals Emitted by a Bottlenose Dolphin when Performing Assisted TherapyMusic Recommender System based on Sentiment Analysis Enhanced with Natural Language Processing Technics author indexsubject indexsearch form
Home Pagealphabetic serial listing  

Services on Demand

Journal

Article

Indicators

Related links

  • Have no similar articlesSimilars in SciELO

Share


Computación y Sistemas

On-line version ISSN 2007-9737Print version ISSN 1405-5546

Comp. y Sist. vol.27 n.1 Ciudad de México Jan./Mar. 2023  Epub June 16, 2023

https://doi.org/10.13053/cys-27-1-3982 

Articles

Feature Selection Ordered by Correlation - FSOC

Arturo Heredia-Márquez1 

Adolfo Guzmán-Arenas1 

Gilberto Lorenzo Martínez-Luna1  * 

11 Instituto Politécnico Nacional, Centro de Investigación en Computación, Mexico. arturoheredia@live.com.mx, aguzman@ieee.org.


Abstract:

Data sets have increased in volume and features, yielding longer times for classification and training. When an object has many features, it often occurs that not all of them are highly correlated with the target class, and that significant correlation may exist between certain pair of features. An adequate removal of “useless” features saves time and effort at data collection, and assures faster learning and classification times, with little or no reduction in classification accuracy. This article presents a new filter type method, called FSOC (Feature Selection Ordered by Correlation), to select, with small computational cost, relevant features. FSOC achieves this reduction by selecting a subset of the original features. FSOC does not combine existing features to produce a new set of fewer features, since the artificially created features mask the relevance of the original features in class assignment, making the new model difficult to interpret. To test FSOC, a statistical analysis was performed on a collection of 36 data sets from several repositories some with millions of objects. The classification percentages (efficiency) of FSOC were similar to other feature selection features. Nevertheless, when obtaining the selected features, FSOC was up to 42 times faster than other algorithms such as Correlation Feature Selection (CFS), Fast Correlation-Based Filter (FCFB) and Efficient feature selection based on correlation measure (ECMBF).

Keywords: Feature selection; data mining; pre-processing; feature reduction; data analysis

1 Introduction

Feature reduction is important in data mining, since its benefits are: simpler and cheaper data collection; less space in memory and disk; smaller processing times, both in training and classification; better visualization and understanding of the results [6].

The reduced set of features should not appreciably decrease the accuracy (percentage of correct answers) of a classifier that uses it.

Moreover, when an object has hundreds of thousands of features, classification becomes complicated, due to ’the course of dimensionality’: as the number of features (dimensions) increase, the data set is very sparsely distributed in the huge spanned dimension space, and many regions of this space are empty [17].

This provokes a curious fact: the distance (for any reasonable metric) between any two points is about the same. Any point is more or less equidistant to all the others [1].

Any three points lie in an almost equilateral triangle. Classifiers that use ’distance’ no longer work well. Such space challenges our intuition about ’closeness’ and ’clustering’, for instance.

Another difficulty arises when data sets are imbalanced, because objects of a certain class are much more abundant than those of other classes.

For instance, sampling the population of a city, the number of people having cancer is quite small compared to the number of people do not having cancer, say, 5% versus 95%.

Many classifiers are tempted to assign to the ’healthy’ class any person, since its error (percentage of incorrect results) will be at most 5%. The main contributions in this paper are:

  • – FSOC, an algorithm that selects a reduced set of features with less computational effort (much less number of comparisons between features to obtain the subset) than other state-of-the-art feature selection algorithms.

  • – The classification accuracy when using this reduced set of features is very similar to the accuracy obtained by using the complete set.

  • – FSOC finds less features than other feature selector algorithms in high dimensional data sets.

  • – FSOC has also good performance for imbalanced data sets.

The paper is organized as follows. Section 1 introduces the reader to the feature selection area. Section 2, Related works, describes relevant previous work. Section 3 describes the FSOC algorithm and its foundation.

Section 4 compares FSOC with other state-of-the-art selection algorithms, using a statistical analysis. Finally, the last Section contains our conclusions and future work.

The algorithms for feature reduction fall in two groups: feature extraction and feature selection. The first group generates new features by combining the original features.

The number of new features is smaller than the number of the original features. Each object in the dataset is now described by the new features.

The new features are a linear or non-linear combination of the original features, and they are used to span a lower dimensional subspace for the original space, or, in recent subspace techniques, for each pattern class.

Unfortunately, these feature extraction or combination algorithms work mainly with only numeric features or categorical, but not both [14], and they present limitations when the classes are highly imbalanced. That is, when the apriori probabilities of the classes are very different. Therefore, they are not suitable for data sets having a mixture of numerical and categorical features. In addition, the resulting features are difficult to explain to a user that seeks to understand why a particular instance was classified in a certain way.

These techniques reformat, transform and combine the features of each object. For this reason, these algorithms were not considered in this paper.

The second group selects a subset of features from the complete set, called ’relevant features’ because they provide information to correctly discriminate the instances with respect to the class; in other words, the features have a correlation with the class [10, 7].

The goal in both groups is to obtain a good set of features, defined by [9] as those that are correlated with the target class but have little or no correlation with each other. The most important methods for feature selection are filter and wrapper methods.

1.1 Filter Methods

The filter methods carry the process of feature selection without the use of any induction (classification) algorithm. They analyze the training data set to obtain statistical characteristics such as the correlation or the degree of association between two features in order to compare and select features with independence of any predictor (classifier algorithm for instance) and association with the class.

These methods are faster than wrapper methods and generalize better because they act independently of the classification algorithm [15].

1.2 Wrapper Methods

Wrapper methods perform feature selection generating candidate subsets of features, and evaluating them by a previously defined classifier.

Because some inductive algorithm (classifier algorithm for instance) is required, their computational cost is greater than filter methods. In addition, the results will be useful mainly for that classification algorithm [21].

1.3 Ranking Methods

These methods use different correlation measures between the features and the target class, producing ordered lists. The method selects those features that have the highest frequency of appearance in the first places of the lists.

However, the limitation of this type of method is that relationships among features are ignored [11]. In general, the search methods eliminate or add features to the set of relevant features, according to certain selection criteria.

The best-known methods are Forward Selection and Backward Selection.

1.4 Forward Selection

This method starts without any feature in the model (the set of selected features), which implies that no previous information of the correlations between features is necessary.

Every feature that is not included in the current model will be validated through some heuristic.

If it adds discriminatory power to the model, it will be included in it. The method continues until no feature provides information, or if all the features are included in the model.

1.5 Backward Selection

This method is completely contrary to the previous model; it starts with all the features included in the model. Then, it determines first the correlations between any two features for next calculations.

Every feature included in the model is considered for its elimination. It will be excluded from the model if its removal increases the discriminatory power of the model and is redundant with other features. The method continues until no feature can be eliminated.

2 Related Work

Most classifiers in data mining have some weakness when the data set has redundant features or features that are not very relevant. In several publications, metrics have been designed to evaluate the relevance of features [4].

Some metrics only work in numeric, nominal or mixed spaces; for example, the Pearson and Spearman correlation work only with numerical data, while the information gain, the symmetric uncertainty coefficient, V Cramer and confusion measure work with nominal features.

CMCD, based on the theory of class separation, relates numerical and mixed features [13]. However, these metrics only allow measuring the correlation between two features, and do not provide information about whether these features have high correlation with the target class.

Different methods for feature extraction and feature selection has been proposed and used in different areas of knowledge, such as the energy sector, the education sector, recommendation systems, among others.

In article [8], a strategy is presented to increase the efficiency of classifying the stress (low, medium and high) of a driver, through the obtaining and analyzing biological changes such as blood pressure, heartbeat, muscle activity, among others.

The process is carried out through two stages; the first consists of obtaining the new features obtained through feature extraction with the intention of reducing ’altered’ measures and characterizing them.

Subsequently, discriminative common vectors are used to generate an identifying vector for drivers and thus classify them. Moreover, for the discriminative vectors, it is necessary to obtain eigenvectors and eigenvalues to transform the space and it could be limited if the matrices are large.

In addition, the explanation of the final result is confusing, because the vectors generated comes from a series of transformations and combinations between features. In [18] some linear and non-linear techniques for the generation of subspaces are explained, which use Cholesky decomposition to create a matrix that approximates the original data.

The uniqueness of these techniques is that they involve Kernel functions to approximate more complex (non-linear) data. Once this matrix is obtained, the training values can be transformed by its orthogonal representation.

Furthermore, these transformations and new representations of the data cause a lesser understanding of the generated model, and do not always help in decision making. In [2], the use of latent factor models in recommendation systems is proposed because of their ease in dealing with scattered matrices (with missing values).

The main idea is to use the high correlations between columns and rows in order to rotate the axes system and eliminate the redundancy in pair wise correlations. These models map the values of the features into a smaller dimensional space and thereby infer recommendation items based on information from other users.

In [3], Principal Components Analysis (PCA) is used to evaluate and obtain an energy sustainability index for rural communities.

Features such as population density, per capita energy consumption, per capita production, proportion of local residents and tourists, among others, were considered. The result shows that a certain region of France has a better energy index than the rest.

A set of new features is obtained by a linear combination of the original features. Two possible drawbacks are: only numerical features can be combined; and the semantics or meaning of this new set is difficult to understand, and the contribution of the original features is not clear.

Instead, the methods of selecting features maintain the meaning of the original features, since they only exclude less relevant features. In [9], the heuristic called merit is proposed to know how ’good’ a subset of features is.

This heuristic takes into account the usefulness of the features individually, while at the same time it measures the level of correlation between them.

In other words, the merit is the ratio of the correlation between the features with the class divided by the correlation of the features with each other.

The Correlation Feature Selection (CFS) algorithm is developed, its results show great efficiency with different data sets with little or no loss of accuracy in classifiers.

The method allows forward and backward searches. The method consists in adding or eliminating a feature that increases merit and ends when merit decreases while removing or adding a feature, or there are no more features to evaluate.

The main limitation in this method is that although not all combinations of features are generated, too many comparisons are made between features, which could be expensive in a high dimensionality data set. Our proposed FSOC algorithm reduces these comparisons, thus running faster.

[20] describe the method Fast Correlation-Based Filter (FCBF), that seeks primarily the answer to two questions: how to decide if a feature is relevant to discriminate instances with respect to the class (a relevance threshold δ is introduced), and how to decide if the relevant feature is redundant with some other feature in the complete feature set.

In order to answer this question, it uses the condition that the correlation between feature A and feature B is greater than or equal to the correlation between feature B and class feature.

The method first excludes from the complete set those non-relevant features, and then, every feature in the remaining subset is compared to all others in the subset, to find if there is a strong relationship between any of them.

In this case, it excludes the feature with less discriminatory power. FCBF has two disadvantages. It computes the correlations of the features with the class to provide an acceptable relevance threshold δ.

It excludes features whose correlation with the class are lower than δ. This is a disadvantage of the algorithm, since a feature that apparently is not very relevant (correlated) with the class, could be useful to avoid loss of classification accuracy (for example, when all the features are required to obtain the greatest discrimination power).

Another disadvantage is that it considers δ and redundancy sequentially. That is, it first filters out those features that are not relevant (have little discriminating power) and then proceeds to eliminate redundant features.

Although it is a very fast algorithm, the efficiency of the classifiers that use the selected subset of features decreases, which implies loss of discriminating power.

[16] defines four groups of features: (1) strongly relevant features, (2) relevant and non-redundant features, (3) relevant but redundant features and (4) weakly relevant and redundant features.

An optimal subset is one that has features of group 1 and group 2. Their ECMBF algorithm exploits these concepts with two parameters, α as the relevance threshold and β as the redundancy threshold.

The first step is to eliminate features that do not comply with α (group 4) and subsequently eliminates those that are redundant (group 3), prevailing those with greater relevance (groups 1 and 2). As in FCBF, considering the parameters α and β in isolation could cause loss of discriminative power.

Moreover, without a previous knowledge of the data set, an initial setting of α and β could be wrong, provoking poor classification accuracy when using the reduced feature set, as compared with using the complete set of features.

In [5] the ANCONE algorithm is developed, that employs the CFS method (Correlation Feature Selection, explained above when describing work [9]), it was used to find personal and socio demographic characteristics associated with the school performance of third grade Mexican High School students in Mathematics.

The complexity of the problem lies in the fact that the data sets to analyze contain approximately 52232 instances (students) and 232 features. Many of these features contain redundant information (Do you have internet at home? Do you have a home computer? Do you have electricity at home?).

From 232 features, 18 were identified as relevant by the CFS method, which are questions about the student’s academic record, the type of school, the educational level of the parents and the student’s academic aspirations. These features increased the efficiency of the classifier from 50% to 68%.

Although the CFS method is effective, when you have a large number of features the algorithm tends to have computationally high costs and requires substantial memory.

Deep learning can be used to extract high-dimensional features, which can be regarded as a complex combination of existing features. This can lead to a reduction of the needed features to accomplish a decent classification.

Therefore, deep learning can be used as a feature reduction method. Nevertheless, it is well known that deep learning takes a long time to converge, especially with massive amount of data having many features. In contrast to this, FSOC is characterized by a short processing time.

3 FSOC Algorithm

The FSOC algorithm uses the heuristic merit (MS) to select those relevant and non-redundant features (optimal subset) [9], which measures how ’good’ a set of features is (see Equation 1), the subset with highest merit will be the optimal subset.

In it, k is the number of features in the optimal subset, rcf is the average correlation between the features and the class feature (target feature), and rff is the average correlations between the features selected.

The same principle is used in test theory to design a composite test for predicting an external variable of interest, ’features’ are individual tests which measure traits related to the variable of interest (class or target feature).

If a group of components increase, it is unlikely that all of them are highly correlated with the target feature and at the same time bear low correlations with each other [19]:

MS=krcfk+k(k1)rff. (1)

3.1 Description of FSOC

Our algorithm (Algorithm 1, below), shows differences with previous state-of-the-art. FSOC starts by producing a set S of the correlations between each dependent feature and the dependent feature (class), and sorting set S in decreasing order.

Algorithm 1 FSOC algorithm 

Set S (initially empty) will contain the features selected as relevant and non-redundant. Subsequently, the first element is extracted (that feature having the highest correlation with the class) from the set S, add it to the set S and save the value of its correlation as the current merit.

Then FSOC evaluates in an orderly manner the inclusion of each element Xi of the set S in S, by comparing the merit (given by Equation 1) of XiS with the current merit.

The feature Xi that induces the highest merit (let us call it m) is added to S and removed from S, as long as that merit m is higher than the current merit. In addition, the current merit is set to m.

The addition of features from S to S continues until the merit of S is greater than the merit of XiS (for any XiS), or there are no more features to analyze. See Algorithm 1.

Two important differences between FSOC and CFS algorithms (Algorithm 2) involve the way in which the features to be included in the optimal subset are searched.

Algorithm 2 CFS algorithm 

The CFS algorithm searches the next feature to be added to the set S (the set of features selected as relevant and non-redundant) in all the set S (the set of features not yet included in S).

If the size of S is large, this repeated search to the complete set S is costly. Instead, FSOC orders once the features in set S by decreasing correlation of each feature with the class feature. The search of the next feature to add to S stops sooner, due to this ordering.

The first difference is found in steps 4 to 9 of the FSOC algorithm, which obtain the correlation of each feature with the class feature and order them in decreasing order.

In algorithm CFS, its first iteration with the features (steps 7 to 20) also repeatedly seeks the feature Xi which obtains the highest merit of XiS, but it does this search without ordering the features by descending correlation with the class.

As it turns out, this step (ordering the features) is fundamental to reduce the computational cost. The second and biggest difference appears in steps 13 to 31 of algorithm 1, where two nested cycles are described.

The first cycle adds feature Xi to set S if the merit of XiS is greater than the merit of S. In other words, it tries to maximize the merit of S.

The second cycle (steps 16 to 26) is internal. Since it searches each feature in decreasing order (of the correlation of the feature with the class), and stops as soon as the next candidate feature Xi fails to have the merit of XiS higher than the merit of S.

It is considered that the other features (“below” Xi in set S) could provide little information because they have less correlation with the class.

Instead, CFS evaluates the merits of all the features of S, and keeps doing so until there is no feature that increases the merit of S.

Due to this early stop, FSOC performs fewer comparisons between features, and therefore fewer correlations between them. It is clear that fewer comparisons between features will produce a lower computational cost: lower CPU usage.

With respect to disk I/O, the training set has to be read once into main memory, either all of it at the same time (if it fits) or in batches of objects (if too large to fit in memory), in order to compute the correlations (lines 4 to 8 in algorithm 1; lines 8 to 16 in algorithm 2).

The pseudocode of algorithm CFS (Algorithm 2) does not make clear whether the merit calculation (line 9 in algorithm 2) causes the complete training set to be read for each feature Xi to be tested, or if some other method is used. In either case, FSOC has a lower or at least equal I/O cost than CFS.

The end result is that FSOC saves total computational cost = CPU time + I/O time. This improvement is very beneficial for data sets with large volumes of information and with a large number of features.

In addition, because of the way FSOC computes set S’ (algorithm 1), the relevant features in S’ are in ascending order of merit. In this manner, it is easy to reduce further the set of relevant features, in the case S’ is too large.

Now, let us compare how FSOC and ECMBF work, the main difference between FSOC and ECMBF are that ECMBF uses two thresholds; α (relevance) and β (redundancy). A poor setting of these thresholds could produce a set S with low classification accuracy.

The search space for these thresholds is two-dimensional in ranges of values in [0-1]; decreasing or increasing them independently does not guarantee that the combination found is ’good’, because the classification accuracy is not necessarily a monotonic function of either of them.

Testing different values of α and β and evaluating their behavior with some predictor (a classifier, for instance) could be impractical. FSOC avoids making comparisons where there is little predictive information, unlike ECMBF where the features that meet the α (relevance) threshold require a second redundancy filter (β), where comparisons between features are unavoidable.

Now, let us compare how FSOC and FCBF work, the main differences between FSOC and FCBF are that, although both algorithms order the features considering their correlation with the class, the two cycles described in steps 13 to 31 of algorithm 1 allow a faster stop and avoid making comparisons (correlations) between features, as opposed to FCBF, where the search is more exhaustive and therefore considers a greater number of comparisons.

In addition, the relevance parameter (δ) is not necessary in FSOC. This is a very important consideration, because poor values assigned to it could cause features to be incorrectly selected and prematurely discarded, resulting in lower precision when sorting with them. Moreover, it is difficult for the user to assign good values to δ.

The next section shows experiments with real data sets where it is observed that FSOC (mainly due to the ordering of features by their correlation with the class and the two nested cycles, already described) helps the reduction in computational costs and number of features selected.

4 Statistical Comparison of FSOC, CFS, ECMBF and FCBF Using Several Classifiers

This section compares FSOC with several feature selection methods, with respect to (1) accuracy (percentage of correct classifications), (2) computational cost (defined as number of necessary comparisons between features), and (3) reduction of features.

The experiments carried out use the accuracy and Kappa measures, since accuracy is an easy measure to understand and although it has a disadvantage when faced with unbalanced data sets, it is complemented by the Kappa measure, that improves on the accuracy measurement by measuring the agreement between predicted and real value, due to chance (the classifier does a random class assignment).

Statistical analysis consists in comparing the algorithms through multiple averages of random executions of data sets.

Our universe U (see Table 1) consists of 36 datasets with nominal, numerical and categorical features, the number of classes ranging from 2 to 26, and the number of instances or objects in the dataset is between 62 and 8,405,0979.

Table 1 Data sets description 

Data set Area Instances # Classes # Features Repository
Nominal Numerical
Adult Social 32561 2 8 6 UCI
Austra Financial 690 2 8 6 UCI
Breast Health 683 2 1 9 UCI
Credit Financial 653 2 9 6 UCI
Default Credit Financial 30000 2 0 23 UCI
Diabetes Health 768 2 0 8 UCI
German Financial 1000 2 13 7 UCI
Glass Physical 214 6 0 9 UCI
Heart Health 303 2 6 7 UCI
Iris Life 150 3 0 4 UCI
Letter Recognition 20000 26 0 16 UCI
Sonar Physical 208 2 0 60 UCI
Wine Chemical 178 3 0 13 UCI
Cardio Health 267 2 0 44 Keel
Coil Identify 9822 2 0 85 Keel
Fars Injury 100968 8 24 5 Keel
Magic Physical 19020 2 0 10 Keel
Ringnorm Physical 7400 2 0 20 Keel
Shuttle Physical 57999 7 0 9 Keel
Spam Computer 4597 2 0 57 Keel
Allaml Biological 72 2 0 7128 scikit-feature
Gli_85 Biological 85 2 0 22283 scikit-feature
Parkinson Health 756 2 0 753 Kaggle
Prostate_ge Biological 102 2 0 5966 scikit-feature
Smk_Can Biological 187 2 0 19993 scikit-feature
Yale Face 165 15 0 1024 scikit-feature
Gisette Digit 7000 2 0 5000 scikit-feature
Leukemia Biological 72 2 0 7070 scikit-feature
Colon Biological 62 2 0 2000 scikit-feature
Madelon Artificial 2600 2 0 500 scikit-feature
Pcmac Text 1943 2 0 3289 scikit-feature
Basehock Text 1993 2 0 4862 scikit-feature
Poker Game 1025010 10 11 0 scikit-feature
Susy Physical 5000000 2 0 18 scikit-feature
Mobile Health Health 1215745 13 0 13 Kaggle
Covid-19 Health 8405079 4 7 0 Kaggle

To reduce the possible bias introduced by a classifier, three classifiers were used in this analysis: A tree classifier (C4.5), an ensemble of tree classifiers (Random Forest), and a Naive Bayes classifier, in such a way that the averages by the central limit theorem normalize the results and they can be compared each other.

The pseudocode of the statistical algorithm used is shown in algorithm 3. For each feature selection algorithm perform randomly select a data set from U (refer to Algorithm 3), then randomly select a classifier (Random Forest, C4.5, Naive Bayes), make the feature selection of the selected data set and classify.

Algorithm 3 Statistical algorithm to compare feature selection algorithms 

Store the obtained value in T until it has at least 36 values. Then, the average of T will be stored in Y, this process will be carried out until the values obtained in Y form a Gaussian distribution.

According to the central limit theorem, a Gaussian distribution will be obtained when there are at least 5 observations in each decile and the normality test (a test to determine whether sample data has been drawn from a normally distributed population [within some tolerance]) has an error of 0.05 (This would be equivalent to say that there is a 5% probability that the distribution is not normal) X25, if these properties are not maintained is necessary return to step 2 (see Algorithm 3).

Comparing the results of FSOC algorithm (Figure 2), when selecting and using the features identified as relevant slightly exceeds the average accuracy than when using the full data set (1.2%), as well as slightly beating the FCBF (3.08%) and ECMBF (1.37%) algorithms.

Fig. 1 Average accuracy percentage of the Random Forest, C4.5 and Naive Bayes classifiers for the 36 data sets, using the different features obtained from the feature selection algorithms. The vertical axis of the graph starts at number 68 to allow a greater appreciation of the results. The Feature Selection Ordered by Correlation (FSOC) algorithm has similar accuracy with Correlation Feature Selection (CFS), but with less features and computational cost, Efficient feature selection based on correlation measure (ECMBF) are slightly better just in Random Forest classifier while Fast Correlation-Based Filter (FCBF) is the lowest 

Fig. 2 Results of the statistical algorithm (Algorithm 3) applied to each feature selector algorithm. Gaussian-shaped histograms are displayed, allowing comparisons between them 

However, CFS is slightly higher than FSOC (0.38%). The difference of average accuracy between FSOC and the rest of the algorithms it is less than 1%, so we could say that it is practically the same or very similar (see Figure 2).

In addition to the previous results, the Chebyshev inequality [12] (see Equation 2) allows us to rank the algorithms by determining their probability that the percentage of correct classifications (y) is in the interval [μkσ,μ+kσ] of their distributions, where μ is the average accuracy, k is the number of standard deviations and σ is the value of the standard deviation:

p(μkσyμ+kσ)11k2. (2)

If we establish a value of k=3.1623 we will know that the values of y will fall in this interval with the probability of p0.9. Therefore, the best (largest) value finding for the feature selection algorithms will be given by μ+kσ.

This is a way of measuring the performance or quality of the algorithm to solve the problems in U. Table 2 shows the algorithm with the corresponding values of μ+kσ, ordered from best to worst accuracy with a probability of falling in the interval (μkσ,μ+kσ) of 0.9.

Table 2 Quality of each algorithm (last column) given by the Chebyshev inequality in descendent order 

Algorithm μ σ μ+kσ
CFS 84.20 1.89 84.20+(k x 1.89) =90.20
FSOC 83.05 1.82 83.05+(k x 1.82) =88.84
ECMBF 82.07 1.93 82.07+(k x 1.93) =88.20
Original 81.91 1.98 81.91+(k x 1.98) =88.18
Data
FCBF 80.47 2.07 80.47+(k x 2.07) =87.02

All the algorithms show good average and upper bound in accuracy (greater than 80). While Table 3 shows the algorithms in order from less to high cost computational. Individual results by data set and classifier are placed in Tables 5, 6 and 7.

Table 3 Computational cost, expressed in number of necessary comparisons between features to obtain the optimal subset 

Algorithm Average Standard deviation
FSOC 2,236.10 798.80
CFS 97,254.09 46,801.22
FCBF 11,088,895.62 6,374,684.55
ECMBF 13,932,895.10 7,728,952.53

Table 4 Number of features selected in each feature selector algorithm for each data set and the computational cost required to obtain them. It is observed that FSOC in the vast majority obtains fewer features than the rest of the algorithms with a smaller number of comparisons. The average is rounded to the whole number 

Data set Original Dara FCBF ECMBF CFS FSOC
#Feat #Feat Cost #Feat Cost #Feat Cost #Feat Cost
Adult 14 7 44 14 92 5 69 5 34
Austra 14 8 42 14 92 1 27 1 16
Breast 10 8 31 8 31 9 45 9 45
Credit 15 8 54 15 104 1 29 1 17
Default 23 8 64 22 237 5 123 5 43
Credit
Diabetes 8 6 22 8 29 3 26 3 17
German 20 9 67 20 191 4 90 4 34
Glass 9 8 29 9 37 5 39 5 34
Heart 13 8 48 13 53 7 76 7 48
Iris 4 3 6 4 7 2 9 2 9
Letter 16 12 77 14 118 9 115 9 79
Sonar 60 51 1263 58 1563 16 884 16 212
Wine 13 11 52 11 60 8 81 6 40
Cardio 44 27 397 41 831 18 665 5 64
Coil 85 61 1918 84 3492 7 652 6 112
Fars 29 8 80 29 407 3 110 3 38
Magic 10 3 18 9 46 3 34 3 19
Ringnorm 20 20 191 20 191 20 210 20 210
Shuttle 9 2 16 8 31 3 30 3 27
Spam 57 16 358 57 1597 10 572 10 131
Allaml 7129 6168 19032070 6818 23276950 28 206335 9 7191
Gli_85 22283 18882 166168130 21750 236834285 53 1201851 10 20045
Parkinson 753 753 283881 753 283881 336 196728 12 879
Prostate_ge 5966 5007 12546176 5264 14899715 24 148850 7 6007
Smk_Can 19993 18229 166168130 19428 188835632 52 1058251 8 20045
Yale 1024 772 299532 831 350156 31 32272 9 1086
Gisette 5000 3504 6152667 4584 10678832 34 174405 7 5047
Leukemia 7070 6713 22538069 7066 24939737 22 162357 5 7094
Colon 2000 1951 1904381 1946 1892599 18 37829 2 2005
Madelon 500 500 125250 500 125250 9 4955 3 509
Pcmac 3289 3170 5027017 3253 5290982 14 49230 10 3354
Basehock 4862 4261 9086273 4815 11608599 13 67977 12 4952
Poker 10 10 54 10 54 4 40 5 30
Susy 18 18 170 18 170 10 143 4 36
Mobile Health 14 14 128 14 128 10 88 10 76
COVID-19 7 7 34 7 34 3 22 1 9
Average 2233 1951 11370464 2153 14417394 22 92922 6 2211

Table 5 Random forest classification for each feature selection algorithm and data set. On Average FSOC is 1.7% approximately below CFS but with less features and computational cost. The results for the SUSY and Poker dataset are incomplete because the model that was built could not be stored in memory 

Data set Original Data FCBF ECMBF CFS FSOC
%Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap.
Adult 84.82 0.566 82.12 0.501 84.82 0.566 85.79 0.574 85.79 0.574
Austra 86.66 0.73 83.76 0.672 86.66 0.73 85.5 0.712 85.5 0.712
Breast 97.21 0.939 97.91 0.942 97.21 0.939 97.91 0.939 97.91 0.939
Credit 86.98 0.738 85.45 0.706 86.98 0.738 86.37 0.729 86.37 0.729
Default_Credit 81.65 0.376 79.07 0.26 86.67 0.369 81.65 0.364 81.65 0.364
Diabetes 76.56 0.474 74.47 0.429 76.56 0.474 72.26 0.379 72.26 0.379
German 75.7 0.352 72.1 0.283 75.7 0.352 69.7 0.246 69.7 0.246
Glass 79.9 0.723 78.5 0.702 79.9 0.723 74.76 0.652 74.76 0.652
Heart 81.18 0.619 79.2 0.578 81.18 0.619 79.86 0.592 79.86 0.592
Iris 95.33 0.93 95.33 0.93 95.33 0.93 94 0.91 94 0.91
Letter 96.46 0.963 96.05 0.959 96.46 0.963 94.78 0.946 94.78 0.946
Sonar 86.05 0.718 83.17 0.658 87.01 0.737 83.65 0.67 83.65 0.67
Wine 98.31 0.975 97.75 0.965 98.31 0.974 97.19 0.958 97.19 0.958
Cardio 80.14 0.233 82.39 0.308 80.89 0.286 81.27 0.339 79.02 0.259
Coil 92.86 0.077 92.5 0.06 92.79 0.073 93.5 0.044 93.95 0.035
Fars 77.77 0.698 78.38 0.704 77.77 0.698 76.68 0.68 76.68 0.68
Magic 87.98 0.728 82.51 0.602 87.68 0.721 82.51 0.602 82.51 0.602
Ring 95.29 0.906 95.06 0.901 95.29 0.906 95.29 0.906 95.29 0.906
Shuttle 99.99 0.999 94.34 0.853 99.97 0.999 99.84 0.996 99.84 0.996
Spam 95.62 0.908 94.69 0.888 95.62 0.908 92.51 0.843 92.51 0.843
Allaml 91.66 0.805 72.22 0.294 88.88 0.735 98.61 0.969 94.4 0.875
Gli_85 84.7 0.591 70.58 0.079 88.23 0.696 97.64 0.944 92.94 0.83
Parkingson 85.31 0.548 87.03 0.606 85.31 0.548 87.56 0.624 86.11 0.596
Prostate_ge 88.23 0.764 81.37 0.626 89.21 0.784 96.07 0.921 94.11 0.882
Smk_can 68.98 0.375 66.84 0.333 59.89 0.2 81.81 0.634 73.79 0.474
Yale 77.57 0.759 75.75 0.74 74.54 0.727 75.75 0.74 64.84 0.623
Gissette 96.9 0.938 92.77 0.855 96.62 0.928 94.88 0.896 87.45 0.749
Leukemia 93.05 0.839 80.55 0.516 93.05 0.842 98.61 0.969 97.22 0.937
Colon 85.48 0.665 77.41 0.462 80.64 0.549 87.09 0.723 88.7 0.74
Madelon 65.84 0.316 65.65 0.313 65.84 0.316 85.57 0.711 71.76 0.435
Pcmac 94.13 0.882 91.71 0.834 94.28 0.885 86.72 0.733 85.33 0.705
Base 98.24 0.964 96.98 0.939 98.49 0.969 90.71 0.814 90.01 0.8
Poker * * * * * * 75.32 0.539 89.87 0.81
Susy * * * * * * * * * *
Mobile Health 95.11 0.91 95.11 0.91 95.11 0.91 95.54 0.905 95.54 0.905
Covid-19 82.24 0.585 82.24 0.585 82.24 0.585 80.93 0.555 79.15 0.53
Average 87.17 0.694 84.146 0.617 86.95 0.688 87.721 0.707 86.017 0.682

Table 6 C4.5 classifications for each feature selection algorithm and data set. On Average FSOC has similar accuracies with respect to CFS, but is better than the complete data set, ECMBF and FCBF 

Data set Original Data FCBF ECMBF CFS FSOC
%Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap.
Adult 85.79 0.586 85.49 0.555 85.79 0.586 85.67 0.566 85.67 0.566
Austra 86.08 0.73 84.05 0.677 86.08 0.73 85.5 0.712 85.5 0.712
Breast 96.04 0.914 95.6 0.904 96.04 0.914 96.04 0.914 96.04 0.914
Credit 85.29 0.703 86.52 0.73 85.29 0.703 86.37 0.729 86.37 0.729
Default_Credit 80.32 0.337 79.55 0.315 86.53 0.345 82.13 0.379 82.13 0.379
Diabetes 73.82 0.416 75 0.438 73.82 0.416 74.6 0.425 74.6 0.425
German 70.7 0.25 71.5 0.266 70.7 0.25 74.6 0.25 74.6 0.25
Glass 65.88 0.541 63.55 0.492 65.88 0.541 65.88 0.652 65.88 0.652
Heart 78.54 0.567 79.86 0.591 78.54 0.567 77.55 0.544 77.55 0.547
Iris 96 0.94 96 0.94 96 0.94 94 0.91 94 0.91
Letter 87.92 0.874 84.59 0.839 87.99 0.875 87.28 0.868 87.28 0.868
Sonar 71.15 0.422 68.75 0.369 71.15 0.422 78.84 0.574 78.84 0.574
Wine 93.82 0.906 88.76 0.829 94.38 0.915 93.82 0.906 92.13 0.88
Cardio 74.9 0.238 79.77 0.307 76.79 0.28 80.52 0.405 78.27 0.256
Coil 93.76 0.002 93.91 0.009 93.94 0.007 94.03 0.006 94.03 0.003
Fars 93.95 0.007 78.47 0.705 93.95 0.007 94.03 0.112 94.03 0.112
Magic 79.85 0.757 81.74 0.574 84.9 0.657 81.74 0.574 81.74 0.574
Ring 85.05 0.661 90.24 0.804 85.05 0.661 90.22 0.804 90.22 0.804
Shuttle 99.97 0.999 94.7 0.862 99.95 0.999 99.81 0.995 99.81 0.995
Spam 92.93 0.852 92.16 0.835 92.93 0.852 91.84 0.827 91.84 0.827
Allaml 88.82 0.754 88.88 0.754 90.27 0.787 90.27 0.795 90.27 0.795
Gli_85 83.52 0.612 87.05 0.691 84.7 0.635 87.05 0.691 89.41 0.747
Parkingson 80.95 0.466 80.95 0.464 80.95 0.466 78.96 0.398 81.61 0.476
Prostate_ge 81.37 0.626 80.39 0.607 85.29 0.705 86.27 0.725 84.31 0.685
Smk_can 60.42 0.195 63.63 0.261 62.56 0.238 68.98 0.378 70.58 0.408
Yale 43.63 0.396 41.81 0.376 44.24 0.402 43.63 0.396 44.84 0.409
Gissette 93.58 0.871 92.06 0.84 93.85 0.877 92.77 0.855 87.3 0.746
Leukemia 91.66 0.816 93.05 0.842 93.05 0.842 94.44 0.875 93.05 0.842
Colon 74.19 0.386 79.03 0.506 82.25 0.573 79.03 0.517 85.48 0.658
Madelon 72.57 0.451 72.57 0.451 72.57 0.451 74.73 0.494 67.61 0.352
Pcmac 82.55 0.65 80.95 0.618 83.47 0.669 80.64 0.611 80.64 0.611
Base 91.21 0.824 87.65 0.753 91.57 0.831 86.55 0.731 86.65 0.733
Poker 64.97 0.346 64.97 0.346 64.97 0.346 73.63 0.504 75.07 0.554
Susy 79.58 0.583 79.58 0.583 79.58 0.583 78.95 0.571 78.01 0.552
Mobile Health 91.31 0.821 91.31 0.821 91.31 0.821 91,39 0.823 91,39 0.823
Covid-19 82.24 0.586 82.24 0.586 82.24 0.586 80.93 0.558 79.15 0.53
Average 81.85 0.57 81.30 0.59 82.29 0.59 83.24 0.60 83.09 0.60

Table 7 Naive Bayes classification for each data set and feature selection algorithm. On average FSOC is slightly below CFS but with less features and computational costs. Also, FSOC is better than full data set, ECMBF and FCBF algorithm. In the Poker data set, an anomaly is shown, where all the feature selectors and the original set do not present any difference, unlike other classifiers (tables 5 and 6). This is probably due to its high unbalance in the classes and that it is not the best classifier for this set 

Data set Original Data FCBF ECMBF CFS FSOC
%Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap. %Acc. Kap.
Adult 83.47 0.501 79.79 0.323 83.47 0.501 79.93 0.327 79.93 0.327
Austra 77.53 0.531 75.5 0.484 77.53 0.531 85.5 0.712 85.5 0.712
Breast 96.04 0.914 96.48 0.924 96.33 0.921 96.33 0.921 96.33 0.921
Credit 78.25 0.55 76.87 0.52 78.25 0.55 86.37 0.729 86.37 0.729
Default_Credit 69.35 0.289 79.63 0.324 61.83 0.22 79.53 0.356 79.53 0.356
Diabetes 76.3 0.466 76.82 0.47 76.3 0.466 76.43 0.455 76.43 0.455
German 75.4 0.381 73.5 0.314 75.4 0.381 74.1 0.319 74.1 0.319
Glass 49.53 0.334 43.92 0.239 49.53 0.334 50 0.332 50 0.332
Heart 82.5 0.646 84.48 0.685 82.5 0.646 83.16 0.685 82.83 0.653
Iris 96 0.94 95.33 0.93 96 0.94 96 0.94 96 0.94
Letter 64.01 0.626 65.49 0.641 65.81 0.644 64.63 0.632 64.63 0.632
Sonar 67.78 0.366 68.75 0.387 69.23 0.394 69.23 0.394 69.23 0.394
Wine 96.62 0.949 97.75 0.965 97.19 0.958 97.19 0.958 97.75 0.965
Cardio 68.53 0.358 67.79 0.33 67.41 0.332 70.78 0.393 72.65 0.407
Coil 78.07 0.121 87.06 0.121 77.86 0.118 93.63 0.057 93.76 0.042
Fars 77.96 0.701 78.06 0.7 77.96 0.701 76.68 0.68 76.68 0.68
Magic 72.68 0.329 76.02 0.441 73.09 0.341 76.02 0.441 76.02 0.441
Ring 97.97 0.959 97.97 0.959 97.97 0.959 97.97 0.959 97.97 0.959
Shuttle 92.8 0.793 94.34 0.853 94.63 0.861 93.47 0.826 93.47 0.826
Spam 79.68 0.604 76.07 0.542 79.68 0.604 86.9 0.716 86.9 0.716
Allaml 98.61 0.969 95.83 0.908 98.61 0.969 98.61 0.969 97.22 0.938
Gli_85 82.35 0.579 78.82 0.49 88.23 0.734 94.11 0.86 92.94 0.833
Parkingson 76.45 0.387 76.58 0.397 76.45 0.387 74.2 0.36 83.2 0.53
Prostate_ge 62.74 0.25 61.76 0.23 60.78 0.784 94.11 0.882 94.11 0.882
Smk_can 60.42 0.211 57.21 0.148 59.89 0.2 77.54 0.548 72.72 0.453
Yale 63.03 0.603 62.42 0.597 62.42 0.597 64.24 0.616 58.78 0.558
Gissette 91.34 0.826 71.77 0.435 91.3 0.826 91.61 0.832 85.38 0.832
Leukemia 90.27 0.779 83.33 0.632 91.66 0.809 98.61 0.969 98.61 0.969
Colon 70.96 0.402 64.51 0.255 64.51 0.255 87.09 0.723 88.7 0.74
Madelon 59.53 0.19 59.53 0.19 59.53 0.19 60.57 0.211 61.8 0.236
Pcmac 80.03 0.601 76.58 0.531 80.13 0.603 78.17 0.561 76.89 0.535
Base 90.01 0.8 83.24 0.664 90.26 0.805 83.19 0.664 81.93 0.639
Poker 50.21 0.203 50.21 0.203 50.21 0.203 50.21 0.203 50.21 0.203
Susy 73.29 0.452 73.29 0.452 73.29 0.452 72.73 0.436 74.58 0.472
Mobile Health 45.91 0.3 45.91 0.3 45.91 0.3 54.93 0.306 54.93 0.306
Covid-19 79.86 0.553 79.86 0.553 79.86 0.553 80.93 0.558 80.93 0.558
Average 76.64 0.541 75.44 0.504 76.51 0.558 80.4 0.598 80.25 0.597

Table 4 shows the number of features selected by each feature selection algorithm. In addition, the number of comparisons necessary to obtain the subset. In addition, individual results by data set and classifier are placed in tables 5, 6 and 7.

Figure 1 compares the relative accuracy from individual experiments with the test data sets, as given by three different classifiers.

The FSOC algorithm gets less features than CFS, while FCBF and ECMBF select in average more features. Algorithms FCBF y ECMBF are the slowest for large features, followed by CFS and finally by FSOC (see Table 3).

It is interesting to see how FSOC behaves with large data sets (large number of instances), such as Poker, Susy, Mobile Health and Covid-19.

It can be seen that FSOC maintains classification efficiency by reducing the features of large data sets, which tend to generate very large models (mainly tree models) that sometimes cannot be stored in main memory.

The same is true for data sets of thousands of features, where very large and poorly understood models tend to be obtained.

Table 8 Pros and cons for each feature selection method 

Method Advantages Disadvantages
Correlation Feature Selection (CFS) Greedy algorithm that obtains an optimal feature set. Maintains and sometimes improves the efficiency of classifiers using selected features compared to using the full set of features. Simple calculations (correlations) that are optimized with matrices. It requires high processing as it needs to compare multiple features, on datasets of thousands of features it is too slow. Feature search is not exhaustive due to the greedy process it uses.
Fast Correlation-Based Filter (FCBF) Search algorithm is fast when the correlation parameter between features and class is high. A redundancy measure is implemented that is obtained directly from the data and is not manipulated by the user. A parameter is required by the user to eliminate irrelevant features (α). An erroneously selected parameter would affect the selected features. Based on the statistical approach described in algorithm 3, the number of selected features far exceeds the CFS and FSOC algorithms
Efficient feature selection based on correlation measure (ECMBF) Implements a new measure to relate nominal and numerical features. It is required to set two parameters; relevance (α) and redundancy (β); assigning these values trivially would imply limited feature selection. Based on the statistical approach described in algorithm 3, the number of selected features far exceeds the CFS and FSOC algorithms.
Feature Selection Ordered by Correlation (FSOC) Fast algorithm for obtaining relevant features. Does not require any assignment of parameters by the user. Simple calculations (correlations) that are optimized with matrices. Ideal for datasets with thousands of features. Greedy search algorithm, does not perform a global search to obtain the best set of relevant features. In datasets with few features, the speed improvement becomes imperceptible. Correlations calculated could be affected by features with extreme values (noise or outliers).

Generating models with fewer features helps improve training and validation times, as well as reducing the space required to store models with little or no loss of predictive information.

In addition, if the reduced features are used for tree models, the tree becomes easier to understand. The algorithm with the lowest average in features and computational cost was FSOC.

Although CFS algorithm very slightly exceeds FSOC on the average of percentage of correct classifications and statistical kappa in the three classifiers, it also far exceeds the average of its computational cost compared to FSOC.

The FCBF algorithm has a low efficiency, it is below the rest of the algorithms, while the use of the complete set of features can cause some classifiers to be confused and overfit.

FSOC allows reducing the features to avoid analyzing statistical relationships that are not very discriminatory for the class (target) or the information contained in them are redundant.

In turn, it helps to eliminate those features that are unreliable because they were apparently answered randomly or based on a non-rational critic, helping to reduce data recovery and maintenance costs.

Could further reduction in time be achieved by parallelization? It is possible to further accelerate FSOC by simultaneously computing the correlation between feature Xi and the class feature C (lines 4 to 7 in algorithm 1), if there were many features.

To achieve this with several processors, the complete training set, as well as a fraction of the features, are given to each of them. Each processor will compute the correlation between the features given to it and the class feature C.

Nevertheless, the rest of the algorithm (lines 10 to 32, algorithm 1) can be run in just one processor, since the time spent by it is short.

5 Conclusions and Future Work

This article presents a method called FSOC that selects relevant features in a way to reduce the computational cost of its selection, with little or no loss of classification accuracy.

Statistical results comparing three well-known methods for feature selection with Feature Selection Ordered by Correlation (FSOC). It measures the computational cost to obtain such reduced set, and the efficiency (number of correct classifications) produced by the selected features.

The efficiency was obtained using classifiers C4.5, Random Forest (decision trees) and Naive Bayes (conditional probability), tested with a collection of 36 data sets available in the open literature.

The results show an efficiency very similar between FSOC and the best algorithm Correlation Feature Selection (CFS), but FSOC is 42 times cheaper with respect to CFS in the computational cost with null or very slight loss of discriminatory power.

Therefore, the FSOC method is especially relevant for high volumes (large data sets) and high dimensionality data (hundreds of thousands of features).

Even though Fast Correlation-Based Filter (FCBF) is fast, it needs to adjust a relevance threshold not to discard useful features. In addition, it classifies with less accuracy, and uses a number of features higher than FSOC does.

The Efficient feature selection based on correlation measure (ECMBF) algorithm is fast, but it is necessary to have a prior knowledge of the data sets, or to find (by trial and error) adequate values for the relevance and redundancy parameters. These extra classifications render it impractical.

Initial work with FSOC on data sets with large amounts of data and high dimensionality (Parkinson, Prostate_ge, Smk_Can, Yale, Gissete, Leukemia, Colon, Madelon, Pcmac, Basehock,Poker, Susy, Mobile Health and Covid-19, Tables 1 and 4, with up to 5,000,000 samples and up to 5,000 features) shows less features selected with no sacrifice in accuracy.

It is planned to perform further testing with additional high dimensionality data sets. It will also be interesting to integrate new ways to discretize numerical features or to find new measures to correlate nominal, numeric and mixed features.

References

1. Blum, A., Hopcroft, J., Kannan, R. (2020). Foundations of Data Science. DOI: 10.1017/9781108755528. [ Links ]

2. Charu, C. A. (2016). Recommender Systems: The Textbook. DOI: 10.1007/978-3-319-29659-3. [ Links ]

3. Doukas, H., Papadopoulou, A., Sawakis, N., Tsoutsos, T., Psarras, J. (2012). Assessing energy sustainability of rural communities using principal component analysis. Renewable and Sustainable Energy Reviews, Vol. 16, No. 4, pp. 1949–1957. DOI: 10.1016/j.rser.2012.01.018. [ Links ]

4. Geng, L., Hamilton, H. J. (2006). Interestingness measures for data mining: A survey. ACM Comput. Surv., Vol. 38, No. 3, pp. 32. DOI: 10.1145/1132960.1132963. [ Links ]

5. Heredia-Márquez, A., Chi-Poot, A. J., Guzmán-Arenas, A., Martínez-Luna, G. L. (2020). ANCONE: An interactive system for mining and visualization of students’ information in the context of planea 2015. Computación y Sistemas, Vol. 24, No. 1. DOI: 10.13053/CyS-24-1-3113. [ Links ]

6. Hernández, D. C. (2018). Metodología de preprocesamiento de datos estructurados para el uso de técnicas de aprendizaje de máquina. Master’s thesis. [ Links ]

7. Hira, Z. M., Gillies, D. F. (2015). A review of feature selection and feature extraction methods applied on microarray data. Advances in bioinformatics, Vol. 2015. [ Links ]

8. Isikli, E. I. (2021). Subspace-based feature extraction on multi-physiological measurements of automobile drivers for distress recognition. Biomedical Signal Processing and Control, Vol. 66, pp. 102504. DOI: 10.1016/j.bspc.2021.102504. [ Links ]

9. Karegowda, A., Manjunath, A., Jayaram, M. A. (2010). Comparative study of attribute selection using gain ratio and correlation based feature selection. International Journal Of Information Technology And Knowledge Management, Vol. 2, pp. 271–277. [ Links ]

10. Khalid, S., Khalil, T., Nasreen, S. (2014). A survey of feature selection and feature extraction techniques in machine learning. 2014 Science and Information Conference, pp. 372–378. DOI: 10. 1109/SAI.2014.6918213. [ Links ]

11. Márquez-Vera, C. (2012). Predicción del fracaso escolar mediante técnicas de minería de datos. IEEE-RITA, Vol. 7, No. 3, pp. 109–117. [ Links ]

12. Navarro, J. (2014). Can the bounds in the multivariate chebyshev inequality be attained?. Statistics and Probability Letters, Vol. 91, pp. 1–5. DOI: 10.1016/j.spl.2014.03.028. [ Links ]

13. Pang-Ning, T., Vipin, K., Jaideep, S. (2002). Selecting the right interestingness measure for association patterns. Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 32–41. DOI: 10.1145/775047.775053. [ Links ]

14. Salkind, N. J. (2007). Encyclopedia of Measurement and Statistics. [ Links ]

15. Sánchez-Maroño, N., Alonso-Betanzos, A., Tombilla-Sanroman, M. (2007). Filter methods for feature selection – a comparative study. Intelligent Data Engineering and Automated Learning - IDEAL, pp. 178–187. DOI: 10.1007/978-3-540-77226-2_19. [ Links ]

16. Sheng-Yi, J., Lian-Xi, W. (2016). Efficient feature selection based on correlation measure between continuous and discrete features. Information Processing Letters, Vol. 116, No. 2, pp. 203–215. DOI: 10.1016/j.ipl.2015.07.005. [ Links ]

17. Solorio-Fernández, S., Carrasco-Ochoa, A., Martínez-Trinidad, J. F. (2022). A survey on feature selection methods for mixed data. Artificial Intelligence Review, Vol. 55, pp. 2821–2846. DOI: 10.1007/s10462-021-10072-6. [ Links ]

18. Teixeira, A. R., Tome, A. M., Lang, E. W. (2009). Feature extraction using linear and non-linear subspace techniques. Artificial Neural Networks – ICANN 2009, pp. 115–124. DOI: 10.1007/978-3-642-04277-5_12. [ Links ]

19. Vanpaemel, W. (2020). Strong theory testing using the prior predictive and the data prior.. Psychological Review, Vol. 127, No. 1, pp. 136–145. DOI: 10.1037/rev0000167. [ Links ]

20. Yu, L., Liu, H. (2003). Feature selection for high-dimensional data: A fast correlation-based filter solution. Proceedings of the Twentieth International Conference on International Conference on Machine Learning, pp. 856–863. [ Links ]

21. Zhang, R., Nie, F., Li, X., Wei, X. (2019). Feature selection with multi-view data: A survey. Information Fusion, Vol. 50, pp. 158–167. DOI: 10.1016/j.inffus.2018.11.019. [ Links ]

Received: May 31, 2021; Accepted: October 16, 2022

* Corresponding author: Gilberto Lorenzo Martínez-Luna, e-mail: lorenzolunacic@gmail.com

Creative Commons License This is an open-access article distributed under the terms of the Creative Commons Attribution License