A Filter-Based Evolutionary Approach for Selecting Features in High-Dimensional Micro-array Data

Abstract

Abstract. Evolutionary algorithms have received much attention in extracting knowledge on high-dimensional micro-array data, being crucial to their success a suitable definition of the search space of the potential solutions. In this paper, we present an evolutionary approach for selecting informative genes (features) to predict and diagnose cancer. We propose a procedure that combines results of filter methods, which are commonly used in the field of data mining, to reduce the search space where a genetic algorithm looks for solutions (i.e. gene subsets) with better classification performance, being the quality (fitness) of each solution evaluated by a classification method. The methodology is quite general because any classification algorithm could be incorporated as well a variety of filter methods. Extensive experiments on a public micro-array dataset are presented using four popular filter methods and SVM.

Ensemble feature selection for high-dimensional data: a stability analysis across multiple domains

“…Note that both SVM and RF classifiers are known to scale well on high-dimensional spaces, but a dimensionality reduction is still crucial, in several domains, for data understanding/interpretability and knowledge discovery purposes. Further, the feature subsets selected through ranking techniques as the one here considered can often be refined through more sophisticated wrapper approaches that further improve the final predictive performance [2,73]. The same AUC analysis is given in Tables 4 and 5, for datasets with higher I/N ratios; here larger values of the cut-off threshold are considered, respectively, th = 10% of N in Table 4 and th = 20% of N in Table 5.…”

A Model for Term Selection in Text Categorization Problems

“…Leveraging on previous studies about tuning GA parameters [11], we set the following values: population size = 30, crossover probability = 1, mutation probability = 0.02, number of generations = 50. Since the GA performs a stochastic search, we considered the results from 3 trials.…”

Evolutionary Feature Selection

“…Support vector machines are applied by many of the studies. Some optimize the resulting parameters in addition to the feature-sets through the evolution [Che+12; Win+11; Din+09; MG+07; Hua+07] and some only use a simple pre-specified SVM [Can+10;Pra+10].…”

“…The proposed algorithm gets similar results using IB5 as the wrapper, slightly higher error rate and fewer features. [Can+10] uses a SVM as fitness function but instead of 10fold cross-validation a leave-one-out-cross-validation is used, meaning for each training example the induction algorithm is tested after construction from the rest. This technique could give more general feature-subsets, but it is also much more expensive.…”