IJCAI’2005, Edinburgh, Scotland August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection Alexey Tsymbal, Padraig Cunningham Department of Computer Science Trinity College Dublin Ireland Mykola Pechenizkiy Department of Computer Science University of Jyväskylä Finland Contents Introduction – Classification and Ensemble Classification Ensemble Feature Selection – strategies – sequential genetic search Our GAS-SEFS strategy – Genetic Algorithm-based Sequential Search for Ensemble Feature Selection Experiment design Experimental results Conclusions and future work IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 2 The Task of Classification J classes, n training observations, p features Given n training instances Training New instance (xi, yi) where xi are values of Set to be classified attributes and y is class CLASSIFICATION Goal: given new x0, predict class y0 Examples: Class Membership of the new instance - prognostics of recurrence of breast cancer; - diagnosis of thyroid diseases; - antibiotic resistance prediction IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 3 Ensemble classification T Learning T1 T2 … TS h1 h2 … hS How to prepare inputs for generation of the base classifiers? (x, ?) Application h* = F(h1, h2, …, hS) (x, y*) IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 4 Ensemble classification T Learning T1 T2 … TS h1 h2 … hS How to combine the predictions of the base classifiers? (x, ?) Application h* = F(h1, h2, …, hS) (x, y*) IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 5 Ensemble feature selection How to prepare inputs for generation of the base classifiers ? – Sampling the training set – Manipulation of input features – Manipulation of output targets (class values) Goal of traditional feature selection – find and remove features that are unhelpful or misleading to learning (making one feature subset for single classifier) Goal of ensemble feature selection – find and remove features that are unhelpful or destructive to learning making different feature subsets for a number of classifiers – find feature subsets that will promote diversity (disagreement) between classifiers EEA IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 6 Search in EFS Search space: 2#Features * #Classifiers Search strategies include: Ensemble Forward Sequential Selection (EFSS) Ensemble Backward Sequential Selection (EBSS) Hill-Climbing (HC) Random Subspacing Method (RSM) Genetic Ensemble Feature Selection (GEFS) Fitness function: Fitnessi acci divi IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 7 Measuring Diversity The fail/non-fail disagreement measure : the percentage of test instances for which the classifiers make different predictions but for which one of them is correct: N 01 N 10 div _ disi,j 11 N N 10 N 01 N 00 The kappa statistic: 1 2 div_ kappai,j 1 2 l Nii 1 i 1 N l N N 2 Ni* N*i i 1 IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 8 Random Subspace Method RSM itself is simple but effective technique for EFS – the lack of accuracy in the ensemble members is compensated for by their diversity – does not suffer from the curse of dimensionality – RS is used as a base in other EFS strategies, including Genetic Ensemble Feature Selection. Generation of initial feature subsets using (RSM) A number of refining passes on each feature set while there is improvement in fitness IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 9 Genetic Ensemble Feature Selection Genetic search – important direction in FS research – GA as effective global optimization technique GA for EFS: – Kuncheva, 1993: Ensemble accuracy instead of accuracies of base classifiers • Fitness function is biased towards particular integration method • Preventive measures to avoid overfitting – Alternative: use of individual accuracy and diversity • Overfitting of individual is more desirable than overfitting of ensemble – Opitz, 1999: Explicitly used diversity in fitness function • RSM for initial population • New candidates by crossover and mutation • Roulette-wheel selection (p proportional to fitness) IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 10 Genetic Ensemble Feature Selection mutation population of genotypes (base classifiers) 10111 10011 01001 01001 10001 00111 11001 01011 recombination f coding scheme selection 10011 10 10001 011 001 01001 01 01011 001 011 phenotype space 10001 10001 01011 11001 x fitness Current ensemble of base classifiers IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 11 Basic Idea behind GA for EFS Ensemble (generation) BC1 init RSM Current Population (diversity) GA BCi New Population (fitness) BCEns. Size Fitnessi acci divi IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 12 Basic Idea behind GAS-SEFS Generation Ensemble BC1 New Population (fitness) init RSM Current Population (accuracies) diversity new BC (fitness) GAi+1 BCi BCi+1 Fitnessi acci divi BCi+1 IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 13 GAS-SEFS 1 of 2 GAS-SEFS (Genetic Algorithm-based Sequential Search for Ensemble Feature Selection) – instead of maintaining a set of feature subsets in each generation like in GA, consists in applying a series of genetic processes, one for each base classifier, sequentially. – After each genetic process one base classifier is selected into the ensemble. – GAS-SEFS uses the same fitness function, but • diversity is calculated with the base classifiers already formed by previous genetic processes • In the first GA process – accuracy only. – GAS-SEFS uses the same genetic operators as GA. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 14 GAS-SEFS 2 of 2 GA and GAS-SEFS peculiarities: – Full feature sets are not allowed in RS – The crossover operator may not produce a full feature subset. – Individuals for crossover are selected randomly proportional to log(1+fitness) instead of just fitness – The generation of children identical to their parents is prohibited. – To provide a better diversity in the length of feature subsets, two different mutation operators are used • Mutate1_0 deletes features randomly with a given probability; • Mutate0_1 adds features randomly with a given probability. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 15 Computational complexity Complexity of GA-based search does not depend on the #features GAS-SEFS: GA: where S is the number of base classifiers, S’ is the number of individuals (feature subsets) in one generation, and Ngen is the number of generations. EFSS and EBSS: where S is the number of base classifiers, N is the total number of features, and N’ is the number of features included or deleted on average in an FSS or BSS search. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 16 Integration of classifiers Selection/Combination Dynamic Static Static Selection (CVM) Weighted Voting (WV) Dynamic Selection (DS) Dynamic Voting with Selection (DVS) Motivation for the Dynamic Integration: Each classifier is best in some sub-areas of the whole data set, where its local error is comparatively less than the corresponding errors of the other classifiers. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 17 Experimental Design Parameter settings for GA and GAS-SEFS: – a mutation rate - 50%; – a population size – 10; – a search length of 40 feature subsets/individuals: • 20 are offsprings of the current population of 10 classifiers generated by crossover, • 20 are mutated offsprings (10 with each mutation operator). – 10 generations of individuals were produced; – 400 (GA) and 4000 (GAS-SEFS) feature subsets. To – – – evaluate GA and GAS-SEFS: 5 integration methods Simple Bayes as Base Classifier stratified random-sampling with 60%/20%/20% of instances in the training/validation/test set; – 70 test runs on each of 21 UCI data set for each strategy and diversity. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 18 GA vs GAS-SEFS on two groups of datasets DVS 0.840 F/N-F disagreement 0.835 Ensemble Size 0.830 3 5 7 10 0.825 0.820 0.815 0.810 GA_gr1 GAS-SEFS_gr1 GA_gr2 GAS-SEFS_gr2 Ensemble accuracies for GA and GAS-SEFS on two groups of data sets (1): < 9 and (2) >= 9 features with four ensemble sizes IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 19 GA vs GAS-SEFS for Five Integration Methods 0.95 Ensemble Size = 10 0.90 0.85 GA GAS-SEFS 0.80 0.75 0.70 0.65 SS WV DS DV DVS Ensemble accuracies for five integration methods on Tic-Tac-Toe IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 20 Conclusions and Future Work Diversity in ensemble of classifiers is very important We have considered two genetic search strategies for EFS. The new strategy, GAS-SEFS, consists in employing a series of genetic search processes – one for each base classifier. GAS-SEFS results in better ensembles having greater accuracy – especially for data sets with relatively larger numbers of features. – one reason – each of the core GA processes leads to significant overfitting of a corresponding ensemble member GAS-SEFS is significantly more time-consuming than GA. – GAS-SEFS = ensemble_size * GA [Oliveira et al., 2003] better results for single FSS based on Pareto-front dominating solutions. – Adaptation of this technique to EFS is an interesting topic for further research. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 21 Thank you! Alexey Tsymbal, Padraig Cunningham Dept of Computer Science Trinity College Dublin Ireland Alexey.Tsymbal@cs.tcd.ie, Padraig.Cunningham@cs.tcd.ie Mykola Pechenizkiy Department of Computer Science and Information Systems University of Jyväskylä Finland mpechen@cs.jyu.fi IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 22 Additional Slides References • • • • [Kuncheva, 1993] Ludmila I. Kuncheva. Genetic algorithm for feature selection for parallel classifiers, Information Processing Letters 46: 163-168, 1993. [Kuncheva and Jain, 2000] Ludmila I. Kuncheva and Lakhmi C. Jain. Designing classifier fusion systems by genetic algorithms, IEEE Transactions on Evolutionary Computation 4(4): 327-336, 2000. [Oliveira et al., 2003] Luiz S. Oliveira, Robert Sabourin, Flavio Bortolozzi, and Ching Y. Suen. A methodology for feature selection using multi-objective genetic algorithms for handwritten digit string recognition, Pattern Recognition and Artificial Intelligence 17(6): 903-930, 2003. [Opitz, 1999] David Opitz. Feature selection for ensembles. In Proceedings of the 16th National Conference on Artificial Intelligence, pages 379-384, 1999, AAAI Press. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 24 GAS-SEFS Algorithm IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 25 Other interesting findings • alpha – were different for different data sets, – for both GA and GAS-SEFS, alpha for the dynamic integration methods is bigger than for the static ones (2.2 vs 0.8 on average). – GAS-SEFS needs slightly higher values of alpha than GA (1.8 vs 1.5 on average). • GAS-SEFS always starts with a classifier, which is based on accuracy only, and the subsequent classifiers need more diversity than accuracy. • # of selected features falls as the ensemble size grows, – this is especially clear for GAS-SEFS, as the base classifiers need more diversity. • integration methods (for both GA and GAS-SEFS): – the static, SS and WV, and the dynamic DS start to overfit the validation set already after 5 generations and show lower accuracies, – accuracies of DV and DVS continue to grow up to 10 generations. IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 26 Paper Summary • New strategy for genetic ensemble feature selection, GASSEFS, is introduced • In contrast with previously considered algorithm (GA), it is sequential; a serious of genetic processes for each base classifier 0.840 0.835 0.830 3 5 7 10 0.825 0.820 0.815 0.810 GA_gr1 GAS-SEFS_gr1 GA_gr2 GAS-SEFS_gr2 • More time-consuming, but with better accuracy • Each base classifier has a considerable level of overfitting with GAS-SEFS, but the ensemble accuracy grows • Experimental comparisons demonstrate clear superiority on 21 UCI datasets, especially for datasets with many features (gr1 vs gr2) IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 27 Simple Bayes as Base Classifier Bayes theorem: P(C|X) = P(X|C)·P(C) / P(X) Naïve assumption: attribute independence P(x1,…,xk|C) = P(x1|C)·…·P(xk|C) If i-th attribute is categorical: P(xi|C) is estimated as the relative freq of samples having value xi as i-th attribute in class C If i-th attribute is continuous: P(xi|C) is estimated thru a Gaussian density function Computationally easy in both cases IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 28 Dataset’s characteristics Data set Instances Classes Balance Breast Cancer 625 286 Car Features 3 2 Categ. 0 9 Num. 4 0 1728 4 6 0 Diabetes 768 2 0 8 Glass Recognition 214 6 0 9 Heart Disease Ionosphere Iris Plants LED LED17 270 351 150 300 300 2 2 3 10 10 0 0 0 7 24 13 34 4 0 0 Liver Disorders 345 2 0 6 Lymphography MONK-1 MONK-2 MONK-3 Soybean Thyroid Tic-Tac-Toe Vehicle Voting Zoo 148 432 432 432 47 215 958 846 435 101 4 2 2 2 4 3 2 4 2 7 15 6 6 6 0 0 9 0 16 16 3 0 0 0 35 5 0 18 0 0 IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 29 GA vs GAS-SEFS for Five Integration Methods 0.95 0.90 Ensemble Size 0.85 3 5 7 10 0.80 0.75 0.70 0.65 SS WV DS DV DVS SS WV DS DV DVS Ensemble accuracies for GA (left) and GAS-SEFS (right) for five integration methods and four ensemble sizes on Tic-Tac-Toe IJCAI’2005 Edinburgh, Scotland, August 1-5, 2005 Sequential Genetic Search for Ensemble Feature Selection by Tsymbal A., Pechenizkiy M.,Cunningham P. 30