Date of Publication

2026

Document Type

Bachelor's Thesis

Degree Name

Bachelor of Science in Statistics Major in Actuarial Science

Subject Categories

Mathematics

College

College of Science

Department/Unit

Mathematics and Statistics Department

Thesis Advisor

Regina M. Tresvalles

Defense Panel Chair

Olivia P. Pagulayan

Defense Panel Member

Kevynn P. Delgado

Abstract (English)

Loan default prediction remains a challenge for financial institutions, including cooperatives, as it affects financial stability and portfolio quality. Using a loan release dataset from a cooperative, this study aims to explore whether a Random Forest and Neural Network algorithm can be effectively used to predict loan defaults in a cooperative setting in the Philippines. These models were compared against the cooperative’s existing logistic regression models for new members, self-employed members, and employed members for loan prediction. The performance of the models were assessed through the Area Under Curve, Gini coefficient, Kolmogorov-Smirnov statistic, and Population Stability Index. Based on the performance metrics, neural network models have the best balance between high predictive power and acceptable stability. Logistic regression shows lower PSI values but its predictive performance is notably lower. The random forest models generally offer improved predictive power over logistic regression but have higher PSI values in several cases. Based on analysis on all three sets of models, the neural network models consistently demonstrate the best overall performance.

Abstract Format

html

Abstract (Filipino)

Ang prediksyon ng loan default ay nananatiling isang hamon para sa mga institusyong pinansyal, kabilang ang mga kooperatiba, dahil nakaaapekto ito sa katatagang pinansyal at kalidad ng kanilang loan portfolio. Gamit ang loan release dataset mula sa isang kooperatiba, layunin ng pag-aaral na ito na tuklasin kung maaaring epektibong gamitin ang mga algorithm na Random Forest at Neural Network sa prediksyon ng pagkabigo sa pagbabayad ng utang mula sa mga kooperatiba sa Pilipinas. Ikinumpara ang mga modelong ito sa kasalukuyang logistic regression models ng kooperatiba para sa mga bagong miyembro, self-employed na miyembro, at mga employed na miyembro sa prediksyon ng loan default. Sinuri ang performance ng mga modelo gamit ang Area Under Curve (AUC), Gini coefficient, Kolmogorov-Smirnov statistic, at Population Stability Index (PSI). Batay sa mga sukatan ng performance, ang mga neural network models ang may pinakamainam na balanse sa pagitan ng mataas na kakayahang mag-predict at katanggap-tanggap na antas ng stability. Ipinakita naman ng logistic regression ang mas mababang PSI values, ngunit kapansin-pansing mas mababa ang predictive performance nito. Ang mga random forest models ay karaniwang may mas mataas na predictive power kumpara sa logistic regression, ngunit may mas mataas na PSI values sa ilang kaso. Batay sa pagsusuri ng tatlong hanay ng mga modelo, ang mga neural network models ang patuloy na nagpapakita ng pinakamahusay na kabuuang performance.

Abstract Format

html

Language

English

Format

Electronic

Keywords

Cooperative societies--Philippines; Loans; Neural networks (Computer science)

Upload Full Text

wf_yes

Embargo Period

4-14-2028

Download

Available for download on Friday, April 14, 2028

Share

COinS