Explainable Machine Learning for Data Extraction Across Computational Social System

This article addresses the explainable machine learning for data extraction on diverse datasets. In many cases, individual or specific approaches have been developed for feature selection (FS) on a certain dataset, but collecting the diversity dataset and demonstrating it through different FS methods are challenging. Thus, this article proposed multiapproaches for FS with the classification of diverse datasets. The proposed framework is developed using various methods, such as extendable particle swarm optimization (PSO), global and local searching, feature ranking, feature clustering, computational cost-based FS, and multiobjective optimization. We effectively used these methods in our proposed work in a single-setting framework. We focused on three essential computational items in our framework: classification accuracy, selected features, and computational times. Due to the diverse dataset, few methods have been considered challenging during computational evaluation for classification accuracy with test cost. We tried to manage the classification accuracy based on total cost and high accuracy with less cost. The proposed framework is experimented with the above methods and analyzed through comparative results on diversity datasets. For example, when regular parameter values are in the range of
$2^{-13}$
–
$2^{-6}$
, the evaluation result affects all items, i.e., decreasing during this range; other values do not affect results. We used thresholds ranging from 0.6 to 0.9 for highly correlated feature pairs as per the support vector machine (SVM) method for recursive feature elimination.