Adaptive fuzzy cluster-guided simple, fast, and efficient feature selection for high-dimensional and highly imbalanced binary-class bioinformatics microarray data

High-dimensional, highly imbalanced microarray data exhibit severe feature redundancy and class overlap, which biases learning toward the majority class and undermines reliable classification. This paper proposes the Adaptive Fuzzy Cluster-Guided Simple, Fast, and Efficient (AFCG-SFE) feature selection model. AFCG-SFE combines two-stage fuzzy feature clustering with mutual information-based intra-cluster refinement to select discriminative features while reducing redundancy. It employs an imbalance-aware penalty-reward fitness function that optimizes F-measure, G-mean, and AUC, and a data-driven penalty-reward mechanism to enhance minority-class sensitivity, penalize redundancy, and reward subsets with stronger feature-label dependency. Additionally, AFCG-SFE enforces a complexity-driven minimum subset size, guided by feature separability (F1) and class overlap (N2), all within a single-agent SFE search. On 20 benchmark datasets, and compared with evolutionary wrapper and non-heuristic baselines, AFCG-SFE achieves the highest or tied-highest classification performance and the lowest train-test Root Mean Square Error (RMSE), while selecting highly reduced feature subsets (FRR > 99%) and achieving lower class overlap (N2) than the evolutionary wrapper baselines.