$k$-Shape Clustering Enhances Group Lasso for Gene Selection and Sample Classification

The surge in high-throughput biological data necessitates efficient tools for knowledge discovery. Group Lasso for logistic regression, a powerful model for sample classification and gene selection in correlated data, relies on robust clustering. This paper addresses the instability of traditional $k$-means variants by introducing $k$-shape clustering into the group Lasso for logistic regression framework. Comparative analyses using simulated and real-world datasets demonstrate that group Lasso for logistic regression with $k$-shape (GLKSH) outperforms $k$-means variants in accuracy and robustness. GLKSH also uniquely identifies informative genes and achieves superior sample classification, highlighting the impact of clustering on group Lasso and offering a valuable approach for gene selection.