Interactive Learning Model For Feature Selection Using Classifiers

Abstract

A lot of attention to feature selection has been focused in the multi-directional field of machine learning to enhance prediction accuracy through feature set analysis and lowering dimensionality. Finding the best features from many feature spaces is challenging, even if several attention techniques have been investigated for feature selection. Therefore, an agent-based interactive learning model is proposed for feature selection with achieving maximum feature subset. At the outset, state-level feature selection is taken in an interactive learning framework, where agents create the environment's based on the state with corresponding actions and build a stable representation of it to feed into interactive learning. The range of feature subset space is explored by agent activity using the interactive learning technique. An interactive learning (IL) model is considered using an exploration or current strategy. The experiment is demonstrated per the proposed model with specific data sets where the suggested strategy significantly improved over more conventional approaches. According to the comparison results, Ada boost performs better (Train score - 1.00). It outperformed competing classifiers on the Parkinson's dataset (testing score:0.93, accuracy:0.93, mean score: 0.67) regarding performance.