Leveraging Feature Selection Algorithms for Early Detection of Type-2 Diabetes

Abstract

Numerous feature selection algorithms have been proposed in the past to solve the curse of dimensionality problem. The choice of apt feature selection algorithm is still a fundamental area of research. Feature selection is a process of identifying and removing irrelevant features and retaining only the highest contributing feature set. Feature selection algorithms are primarily used in applications like healthcare where the classification accuracy needs to be high. This study proposes three simple algorithms namely Mean based Feature Selection (MbFS), Correlation based Feature Selection (CbFS) and Recursive Elimination based Feature Selection (REbFS) for feature selection that efficiently predict early type-2 diabetes. The proposed algorithms are compared with Greedy Stepwise Search Feature Selection algorithm. Exploiting multilayer perceptron classifier, the mean based feature selection yields an accuracy of 80.1 and 79.2%, the correlation based feature selection yields an accuracy of 80.5 and 80.5%, and the recursive elimination based feature selection yields an accuracy of 64.9 and 75% for 768 instances and 2000 instances respectively.

Keywords: Feature selection, mean based, correlation, recursive elimination, type 2 diabetes

Cite this Article M. Lincy, A. Meena Kowshalya. Leveraging Feature Selection Algorithms for Early Detection of Type-2 Diabetes. Journal of Computer Technology & Applications. 2020; 11(1): 13–20p.