Machine Learning using Multi-Objective Evolutionary Algorithms

Abstract

Broadly speaking, machine learning consists of handling a large amount of data.The the quality of these data affect so much the accuracy of the learning modelwhatever the performance of the employed learning algorithm. Therefore, atechnique should be invoked to improve the representation of the dataset.Feature selection try to offer to the learning algorithm well-represented datasetby removing irrelevant and redundant features and selecting the most infor-mative features. This act results, mainly, in decreasing the number of featuresand improving the prediction accuracy of the learning algorithmn. However,the conflicting design between number/accuracy makes feature selection amulti-objective problem. Therefore, it is more suitable to treat such as situationby using a multi-objective optimization algorithm rather than single-objectiveapproach. Consequently, we propose in this thesis, two evolutionary computationalgorithms for solving multi-objective optimization problems in general manner,and for tackling feature selection problem.The first algorithm called Guided Population Archive Whale Optimization Al-gorithm ”GPAWOA”. The proposed algorithm represents a viable alternative forsolving multi-objective optimization problems. It uses the notion of Pareto domi-nance to compare between the candidate solutions, adopts an external archive tomaintain the elitism concept and guide the population towards promising regionswithin the search space, and employed the computation of the crowding distanceto improve the distribution of solutions.The second algorithm investigates GPAWOA for addressing feature selection inclassification problem. The proposed algorithm, namely FW-GPAWOA, employsa transfert function to make it able to deal with discrete problems, and combinesfilter and wrapper models into a single system to benefits from each model’smerits in order to reduce the feature set cardinality and improve the predictionaccuracy of the learning algorithm.