Analisis Perbandingan Algoritma Random Forest, Xgboost, dan Regresi Logistik dalam Memprediksi Keterlambatan Penyelesaian Proyek Riset Nasional

Setyawati, Dian (2026) Analisis Perbandingan Algoritma Random Forest, Xgboost, dan Regresi Logistik dalam Memprediksi Keterlambatan Penyelesaian Proyek Riset Nasional. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032222132-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Proyek riset nasional, khususnya Program Riset Inovatif Produktif (RISPRO) Lembaga Pengelola Dana Pendidikan (LPDP), merupakan inisiatif strategis yang kerap menghadapi risiko keterlambatan penyelesaian proyek riset. Keterlambatan tersebut berpotensi menimbulkan kerugian finansial maupun strategis, sehingga diperlukan pendekatan berbasis data untuk memahami penyebabnya sekaligus memprediksi kemungkinannya. Penelitian ini bertujuan untuk mengidentifikasi faktor-faktor yang mempengaruhi keterlambatan penyelesaian proyek riset dan membandingkan kinerja algoritma Random Forest, XGBoost, dan Regresi Logistik dalam memprediksi keterlambatan tersebut. Data yang digunakan berupa data historis proyek RISPRO LPDP yang telah berstatus selesai pada periode 2013 hingga Februari 2026. Identifikasi faktor-faktor yang berpengaruh terhadap keterlambatan dilakukan melalui triangulasi uji statistik (Mann-Whitney U dan Chi-Square), koefisien serta odds ratio pada Regresi Logistik, dan kepentingan fitur pada model berbasis pohon. Hasil penelitian menunjukkan bahwa determinan keterlambatan yang paling konsisten lintas-metode adalah durasi proyek (peningkatan peluang terlambat 3,3 kali lipat per simpangan baku), besaran pendanaan dengan pola non-linier berbentuk huruf U di mana proyek berdana menengah Rp2–5 miliar paling berisiko (hingga 4,7 kali lipat), keterlambatan pencairan dana tahap kedua, dan penjadwalan monitoring yang belum tepat waktu. Random Forest ditetapkan sebagai model prediksi terbaik berdasarkan F1-Score (0,907) dan recall tertinggi (0,942).Temuan ini memberikan dasar bagi LPDP untuk mengefisienkan alokasi sumber daya pendampingan dan memperbaiki proses pengelolaan proyek riset secara berbasis data.
======================================================================================================================================
National research projects, particularly the Innovative Productive Research (RISPRO) Program managed by the Indonesia Endowment Fund for Education (LPDP), are strategic initiatives that frequently face the risk of project completion delays. Such delays potentially result in financial and strategic losses, necessitating a data-driven approach to understand the underlying causes and predict their likelihood. This study aims to identify the key factors influencing research project completion delays and compare the performance of Random Forest, XGBoost, and Logistic Regression algorithms in predicting these delays within the Indonesian research funding context. The dataset comprises historical data from RISPRO LPDP projects completed between 2013 and February 2026. The identification of significant factors contributing to delays was conducted through a triangulation of statistical tests (Mann-Whitney U and Chi-Square), coefficients and odds ratios from Logistic Regression, and feature importance analysis from tree-based models. The results indicate that the most consistent determinants of delay across methods are project duration (which increases the odds of delay by 3.3 times per standard deviation), the funding amount—which exhibits a non-linear U-shaped pattern where medium-funded projects between IDR 2–5 billion carry the highest risk (up to 4.7 times)—delays in the disbursement of the second-phase funding, and untimely monitoring schedules. Random Forest was determined to be the optimal predictive model, achieving the highest F1-Score (0.907) and recall (0.942). These findings provide a solid foundation for LPDP in Indonesia to optimize the allocation of assistance resources and improve research project management through a data-driven approach.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	machine learning, predictive model, LPDP RISPRO, project delay, project management, random forest, xgboost, logistic regression, keterlambatan proyek, manajemen proyek, model prediktif
Subjects:	L Education > L Education (General) T Technology > T Technology (General) > T56.8 Project Management T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Dian Setyawati
Date Deposited:	03 Aug 2026 02:18
Last Modified:	03 Aug 2026 02:22
URI:	http://repository.its.ac.id/id/eprint/141974

Actions (login required)

View Item