Interpretable Deep Forest untuk Analisis High-Dimensional pada Data Genomik (Studi Kasus: Klasifikasi Tingkat Keparahan Kanker Prostat)

Farras, Azhar (2026) Interpretable Deep Forest untuk Analisis High-Dimensional pada Data Genomik (Studi Kasus: Klasifikasi Tingkat Keparahan Kanker Prostat). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003242011-Azhar Farras_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Penelitian ini mengevaluasi algoritma deep forest sebagai metode ensemble learning berbasis pohon keputusan untuk klasifikasi biner kanker prostat dengan data genomik berdimensi tinggi. Data yang digunakan berupa profil metilasi DNA dari repositori TCGA PRAD yang dipetakan oleh Siena Clinical, PT Siena Sains Medika, di mana sinyal fluoresensi ditransformasi menjadi metrik M value skala kontinu. Untuk mengatasi kutukan dimensi, seleksi fitur dilakukan menggunakan metode baseline random forest. Mekanisme cascade pada deep forest diimplementasikan sebagai encoder otomatis untuk menangkap interaksi non-linear yang kompleks secara stabil. Hasil pengujian ini menungjukkan bahwa arsitektur deep forest pada subset fitur optimal (Top-100) secara mutlak mengungguli model random forest, dengan capaian batas performa optimal pada tingkat akurasi 0,892, AUC 0,937, sensitivitas 0,783, spesifisitas 0,933, dan F1-Score 0,800. Untuk mengatasi sifat black-box dari arsitektur model, penelitian ini mengintegrasikan motode SHAP (SHaplye Additive exPlanations). Hasil interpretasi SHAP berhasil mengidentifikasi gen NENF, NFIC, dan PTMS sebagai tiga biomarker epigenetik utama penentu klasifikasi, yang validitas klinisnya terkait proliferasi dan sistem imun tumor telah terkonfirmasi melalui validasi komputasional. Hasil analisis ini membuktikan bahwa deep forest mampu menghasilkan klasifikasi penyakit secara presisi sekaligus memberikan transparansi diagnostik yang relevan bagi pengembangan bioinformatika.
===========================================================
===============================================================================
This study evaluates the Deep Forest algorithm as a decision tree-based ensemble learning method for the binary classification of prostate cancer using high-dimensional genomic data. The data used consists of DNA methylation profiles from the TCGA PRAD repository mapped by Siena Clinical, PT Siena Sains Medika, where fluorescence signals are transformed into continuous-scale M-value metrics. To address the curse of dimensionality, feature selection is performed using the baseline Random Forest method. The cascade mechanism in Deep Forest is implemented as an automatic encoder to stably capture complex nonlinear interactions. The test results show that the deep forest architecture on the optimal feature subset (Top-100) absolutely outperforms the random forest model, achieving optimal performance thresholds at an accuracy of 0.892, AUC of 0.937 sensitivity of 0.783, specificity of 0.933, and an F1-Score of 0.800. To address the black-box nature of the model architecture, this study integrated the SHAP (SHapley Additive exPlanations) method. The SHAP interpretation results successfully identified the NENF, NFIC, and PTMS genes as the three primary epigenetic biomarkers determining classification, whose clinical validity regarding tumor proliferation and the immune system has been confirmed through computational validation. The results of this analysis demonstrate that deep forests are capable of accurately classifying diseases while providing mechanistic diagnostic transparency that is relevant to the development of bioinformatics.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Forest, Metilasi DNA, SHAP, Kanker Prostat Biomarker Deep Forest, DNA Methylation, SHAP, Prostate Cancer Biomarker.
Subjects:	Q Science Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics Q Science > QA Mathematics > QA278.2 Regression Analysis. Logistic regression Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QH Biology > QH426 Genetics Q Science > QR Microbiology > QR 201.T84 Tumors. Cancer
Divisions:	Faculty of Mathematics and Science > Statistics > 49101-(S2) Master Thesis
Depositing User:	Azhar Farras
Date Deposited:	05 Aug 2026 01:15
Last Modified:	05 Aug 2026 01:15
URI:	http://repository.its.ac.id/id/eprint/143701

Actions (login required)

View Item