Prediksi Survivabilitas Pasien Kanker Paru-Paru Menggunakan Random Survival Forest Dengan Seleksi Fitur Genetik Berbasis Cox-Lasso

Phalosa, Reinasya Diar (2026) Prediksi Survivabilitas Pasien Kanker Paru-Paru Menggunakan Random Survival Forest Dengan Seleksi Fitur Genetik Berbasis Cox-Lasso. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221131-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (17MB) | Request a copy

Abstract

Kanker paru-paru merupakan kanker yang menduduki peringkat pertama tertinggi baik untuk penyebab kasus kanker baru maupun penyebab kematian akibat kanker secara global. Salah satu solusi untuk meningkatkan angka kesintasan pasien kanker paru-paru adalah dengan prediksi survival, dimana penelitian ini membandingkan kinerja dari tiga model survival yaitu Random Survival Forest (RSF), Cox-Proportional Hazards (Cox-PH), dan Gradient Boosting Survival Analysis (GBSA) dalam memprediksi tingkat kelangsungan hidup pasien kanker paru-paru berdasarkan fitur genetik. Gen yang dianggap paling berpengaruh terhadap prediksi kelangsungan hidup ditentukan melalui tiga tahapan seleksi fitur berurutan, yaitu High Variance Filter, Univariate Cox Filter, dan Cox-LASSO, yang mereduksi dimensi data dari 60660 gen menjadi 19 gen terpilih. Hasil seleksi fitur ini selanjutnya diringkas menjadi skor risiko individual yang diintegrasikan dengan fitur klinis pasien untuk konstruksi model survival. Hasil evaluasi pada data uji menunjukkan bahwa GBSA dengan kombinasi Skor Risiko dan Fitur Klinis menghasilkan kinerja terbaik secara keseluruhan, dengan Time-Dependent AUC pada titik waktu 3 tahun sebesar 0.8490, Brier Score sebesar 0.1241, dan C-Index sebesar 0.7365, sementara RSF dengan kombinasi fitur yang sama menghasilkan Time-Dependent AUC sebesar 0.8297 dan Cox-PH sebesar 0.8269. Pengujian lebih lanjut pada data validasi menunjukkan bahwa Cox-PH memiliki kestabilan kinerja yang paling baik antar data uji dan data validasi, sementara GBSA mengalami penurunan kinerja yang lebih besar. Interpretasi SurvSHAP(t) menunjukkan bahwa gen-gen seperti DKK1, LINC01322, dan TCN1 secara konsisten berada di antara prediktor genomik terkuat pada ketiga model.
=================================================================================================================================
Lung cancer is the leading cause of both new cancer cases and cancer deaths globally. One solution for improving lung cancer patient survival rates is survival prediction, and this study compares the performance of three survival models namely Random Survival Forest (RSF), Cox-Proportional Hazards (Cox-PH), and Gradient Boosting Survival Analysis (GBSA) for predicting lung cancer patient survival based on genetic features. Genes considered to have the greatest influence on survival prediction were determined through three sequential feature selection stages: High Variance Filter, Univariate Cox Filter, and Cox-LASSO, which reduced the data dimensionality from 60,660 genes to 19 selected genes. The results of this feature selection were then summarized into an individual risk score that was integrated with the patient's clinical features for survival model construction. The evaluation results on the test data showed that GBSA with a combination of Risk Score and Clinical Features produced the best overall performance, with a Time-Dependent AUC at the 3-year time point of 0.8490, a Brier Score of 0.1241, and a C-Index of 0.7365, while RSF with the same feature combination produced a Time-Dependent AUC of 0.8297 and Cox-PH of 0.8269. Further testing on the validation data showed that Cox-PH had the best performance stability between the test and validation data, while GBSA experienced a greater decline in performance. Interpretation of SurvSHAP(t) showed that genes such as DKK1, LINC01322, and TCN1 were consistently among the strongest genomic predictors in all three models.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kanker Paru-Paru, Prediksi Survival, Cox-LASSO, Cox-PH, RSF, GBSA, Lung Cancer, Survival Prediction, Cox-LASSO, Cox-PH, RSF, GBSA
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Reinasya Diar Phalosa
Date Deposited:	27 Jul 2026 03:12
Last Modified:	27 Jul 2026 03:12
URI:	http://repository.its.ac.id/id/eprint/137546

Actions (login required)

View Item