The Fisher Component-based Feature Selection Method
Abdul Baseer Buriro, Santosh Kumar · Engineering Technology & Applied Science Research · 2022
A feature selection technique is proposed in this paper, which combines the computational ease of filters and the performance superiority of wrappers. The technique sequentially combines Fisher-score-based ranking and logistic regression-based wrapping. On synthetically generated data, the 5-fold cross-validation performances of the proposed technique were compatible with the performances achieved through Least Absolute Shrinkage and Selection Operator (LASSO). The binary classification performances in terms of F1 score and Geometric Mean (GM) were evaluated over a varying imbalance ratio of 0.1:0.9 – 0.5:0.5, a number of informative features of 1 – 30, and a fixed sample size of 5000.