Data-driven based Optimal Feature Selection Algorithm using Ensemble Techniques for Classification

Abstract

The shift in paradigm with advanced Machine Learning algorithms will help to face the challenges such as computational power, training time, and algorithmic stability. The individual feature selection techniques, hardly give the appropriate feature subsets, that might be vulnerable to the variations induced at the input data and thus led to wrong conclusions. An expedient technique should be designed for approximating the feature relevance to improve the performance for the data. Unlike the prevailing techniques, the novelty of the proposed Data-driven based Optimal Feature Selection (DOFS) algorithm is the optimal k-value ‘kf’ determined by the data for effective feature selection that minimizes the computational complexity and expands the prediction power using the gradient descent method. The experimental analysis of proposed algorithm is demonstarted with ensemble techniques for the non-communicable disease such as diabetes mellitus dataset produces an accuracy of 80.80%, whereas comparative performance analysis for benchmark dataset depicts the improved accuracy of 86.03%.