Perbandingan Prediksi Loan at Risk (LAR) dengan Pendekatan Langsung dan Kolektibilitas Menggunakan Metode Long Short-Term Memory (LSTM)

Fajrin, Jihan Fitra (2025) Perbandingan Prediksi Loan at Risk (LAR) dengan Pendekatan Langsung dan Kolektibilitas Menggunakan Metode Long Short-Term Memory (LSTM). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006211071-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Bank berperan penting dalam perekonomian nasional dengan menyalurkan dana dari masyarakat dalam bentuk kredit. Risiko kredit, yang selama ini diukur menggunakan Non-Performing Loan (NPL), merupakan tantangan utama dalam operasional bank. Namun, NPL memiliki keterbatasan dalam menggambarkan potensi risiko kredit yang lebih luas, sehingga digunakan indikator alternatif yang lebih komprehensif, yaitu Loan at Risk (LaR). Penelitian ini bertujuan untuk mengembangkan model prediksi risiko kredit berbasis LaR menggunakan metode Long Short-Term Memory (LSTM), yang merupakan jenis Recurrent Neural Network (RNN) yang unggul dalam mengolah data runtun waktu dan mengenali pola jangka panjang. Model dibangun menggunakan dua pendekatan: pendekatan langsung yang memprediksi nilai LaR berdasarkan data historisnya, dan pendekatan tidak langsung yang menghitung prediksi LaR berdasarkan hasil prediksi masing-masing kolektibilitas (KOL 1 hingga KOL 5). Data yang digunakan berupa data bulanan kolektibilitas kredit bank umum konvensional di Indonesia pada periode Juli 2014 hingga Juni 2024 dalam satuan miliar rupiah. Proses pemodelan meliputi analisis stasioneritas, pemilihan input berdasarkan lag signifikan dari PACF, serta pelatihan dan pengujian model. Hasil penelitian menunjukkan bahwa pendekatan langsung memberikan hasil prediksi yang lebih akurat, dengan model terbaik terdiri atas 13 input neuron, 4 hidden neuron, dan 1 output neuron, serta nilai Mean Absolute Error (MAE) sebesar 0,001134. Pendekatan kolektibilitas memberikan nilai MAE sebesar 0,001777, dengan lag ke-10 dan lag ke-12 menjadi input signifikan yang paling sering muncul. Hasil peramalan untuk 6 periode ke depan menunjukkan kecenderungan penurunan LaR yang meskipun tidak signifikan, dapat menjadi sinyal positif bagi penguatan strategi mitigasi risiko kredit dan perencanaan bisnis perbankan ke depan.
=======================================================================================================================================
Banks play a vital role in the national economy by channeling public funds into credit. Credit risk, commonly measured using Non-Performing Loan (NPL), remains a key operational challenge for banks. However, NPL has limitations in capturing broader credit risk potential, prompting the use of a more comprehensive alternative indicator: Loan at Risk (LaR). This study aims to develop a credit risk prediction model based on LaR using the Long Short-Term Memory (LSTM) method, a type of Recurrent Neural Network (RNN) well-suited for processing time series data and identifying long-term patterns. The model is constructed using two approaches: a direct approach that predicts LaR based on its historical data, and an indirect approach that estimates LaR from the predicted values of each credit collectibility level (KOL 1 to KOL 5). The dataset comprises monthly credit collectibility data from conventional commercial banks in Indonesia for the period of July 2014 to June 2024, measured in billion rupiahs. The modeling process involves stationarity analysis, input selection based on significant lags from PACF, followed by training and testing of the model. The results show that the direct approach yields better prediction performance, with the best LSTM model consisting of 13 input neurons, 4 hidden neurons, and 1 output neuron, achieving a Mean Absolute Error (MAE) of 0.001134. The collectibility-based approach yields an MAE of 0.001777, with lag 10 and lag 12 appearing most frequently as significant inputs. The six-period forecast indicates a downward trend in LaR, although not significant, which can serve as a positive signal for enhancing credit risk mitigation strategies and future business planning in the banking sector.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Deret Waktu, Kolektibilitas, Loan at Risk (LaR), Long Short-Term Memory (LSTM), Risiko Kredit Collectability, Credit risk, Loan at Risk (LaR), Long Short-Term Memory (LSTM), Time-Series Analysis
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HG Finance > HG3751 Credit--Management. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Jihan Fitra Fajrin
Date Deposited:	29 Jul 2025 08:15
Last Modified:	29 Jul 2025 08:15
URI:	http://repository.its.ac.id/id/eprint/122860

Actions (login required)

View Item