AN EVALUATION OF LASSO-MCGA FOR BIOMARKER SELECTION IN CANCER CLASSIFICATION USING MICROARRAY TABULAR DATA

KAVINDRA ALAIKA AHMAD, Aina Musdholifah, S.Kom., M.Kom., Ph.D

2026 | Skripsi | ILMU KOMPUTER

Penelitian ini mengevaluasi sebuah metode hybrid gene selection, LASSO-mCGA, pada empat dataset microarray kanker: colon, diffuse large B-cell lymphoma (DLBCL), lung, dan prostate. LASSO-mCGA menggabungkan least absolute shrinkage and selection operator (LASSO) untuk dimensionality reduction dengan modified compact genetic algorithm (mCGA) untuk seleksi biomarker menggunakan K-nearest neighbors (KNN), Naive Bayes (NB), Random Forest (RF), dan Support Vector Machine (SVM) sebagai fitness evaluator. LASSO mengurangi jumlah gen dari beberapa ribu menjadi puluhan atau ratusan. SVM merupakan classifier berkinerja terbaik pada subset gen yang dikurangi. mCGA menyempurnakan subset ini untuk mendapatkan kumpulan biomarker akhir yang terdiri dari 3 hingga 103 gen dan mencapai akurasi 100% di semua dataset. Jika dibandingkan dengan metode hybrid gene selection yang lain, LASSO-mCGA mencapai atau menyamai kinerja klasifikasi tertinggi, meskipun dengan set biomarker yang lebih besar. Hasil ini menunjukkan bahwa LASSO-mCGA adalah metode hybrid gene selection yang kompetitif untuk klasifikasi kanker.

This research evaluates the performance of a hybrid gene selection framework, LASSO-mCGA, on four benchmark cancer microarray datasets: colon, diffuse large B-cell lymphoma (DLBCL), lung, and prostate. The framework combines the least absolute shrinkage and selection operator (LASSO) for dimensionality reduction with the modified compact genetic algorithm (mCGA) for biomarker selection using K-nearest neighbors (KNN), Naive Bayes (NB), Random Forest (RF), and Support Vector Machine (SVM) as fitness evaluators. LASSO reduced the number of genes from several thousands to tens or low hundreds while preserving high discriminative power, with SVM emerging as the best-performing classifier on the reduced gene subsets. mCGA further refined these subsets to obtain the final set of biomarkers consisting of 3 to 103 genes and achieved 100?curacy across all datasets. When compared with existing hybrid gene selection methods, LASSO-mCGA attained or matched the highest classification performance, albeit with larger biomarker sets. The results indicate that LASSO-mCGA is a competitive hybrid gene selection method for cancer classification.

Kata Kunci : Cancer classification, Gene expression data, Biomarker selection, LASSO, Modified compact genetic algorithm (mCGA)