Penerapan Clustering-based Customer Representation Learning Untuk Prediksi Persetujuan Kredit Berdasarkan Karakteristik Debitur

Wardani, Lola Remitha Septya (2015) Penerapan Clustering-based Customer Representation Learning Untuk Prediksi Persetujuan Kredit Berdasarkan Karakteristik Debitur. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211023-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Persetujuan kredit dilakukan secara selektif untuk meminimalkan risiko gagal bayar. Oleh karena itu, memahami karakteristik debitur secara mendalam menjadi langkah penting dalam menilai kelayakan kredit. Salah satu upaya untuk mendukung proses ini adalah memprediksi kemungkinan persetujuan pinjaman berdasarkan karakteristik debitur. Supaya prediksi dapat dilakukan dengan baik, keseluruhan data karakteristik debitur yang beragam perlu disajikan dalam bentuk representasi numerik yang terstruktur. Penelitian ini menerapkan pendekatan Clustering-based Representation Learning untuk membentuk representasi debitur dari data statis yang terdiri dari berbagai atribut. Metode ini menangkap pola tersembunyi dengan mempertimbangkan kemiripan antar debitur dalam ruang fitur, sehingga tetap efektif meskipun tanpa data transaksi. Representasi dibentuk melalui fungsi agregasi pada fitur kategorikal, perhitungan jarak cosine terhadap profil kategori dan pusat cluster. Proses clustering menggunakan algoritma K-Medoids menghasilkan tiga segmen debitur yang mencerminkan tingkat risiko kredit yang berbeda. Representasi tersebut dikombinasikan dengan fitur asli dan digunakan sebagai input dalam model klasifikasi Random Forest. Hasil evaluasi menunjukkan bahwa model mencapai akurasi sebesar 93% dan F1-score sebesar 88,95%. Analisis feature importance menunjukkan bahwa riwayat kredit macet, suku bunga, jarak terhadap profil kepemilikan rumah, dan pendapatan tahunan menjadi fitur paling berpengaruh. Temuan ini menunjukkan bahwa representasi berbasis cluster memberikan akurasi prediksi cukup tinggi, serta mendukung pengambilan keputusan kredit berdasarkan karakteristik debitur.

=========================================================================================================================================

Credit approval is conducted selectively to minimize the risk of default. Therefore, understanding the characteristics of borrowers in depth becomes an essential step in assessing creditworthiness. One approach to support this process is to predict the likelihood of loan approval based on borrower characteristics. To enable accurate predictions, the diverse data on borrower characteristics needs to be presented in a structured numerical representation. This study applies a Clustering-based Representation Learning approach to construct borrower representations from static data consisting of various attributes. This method captures hidden patterns by considering the similarity between borrowers in the feature space, allowing it to remain effective even in the absence of transactional data. The representations are formed through aggregation functions on categorical features, followed by cosine distance calculations to category profiles and cluster centers. The clustering process, performed using the K-Medoids algorithm, produces three borrower segments that reflect different levels of credit risk. These representations are then combined with the original features and used as input to a Random Forest classification model. Evaluation results show that the model achieved an accuracy of 93% and an F1-score of 88.95%. Feature importance analysis reveals that credit default history, interest rate, distance to homeownership profile, and annual income are the most influential features. These findings indicate that the cluster-based representation provides sufficiently high predictive accuracy and supports credit decision making based on borrower characteristics.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi Persetujuan Pinjaman, Representasi Debitur, Clustering, K-Medoids, Random Forest. Loan Approval Prediction, Debtor Representation, Clustering, K-Medoids, Random Forest.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA9.58 Algorithms T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Mathematics and Science > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Lola Remitha Septya Wardani
Date Deposited:	01 Aug 2025 06:33
Last Modified:	01 Aug 2025 06:33
URI:	http://repository.its.ac.id/id/eprint/125201

Actions (login required)

View Item