Credit Risk Analysis Using C5.0 Decision Tree

Author: Zaki Abiyu Aqilah
Tools: R (openxlsx, C50, reshape2)
Goal: Predict credit risk rating (1–5) based on customer profile

---

Project Overview

This project builds a C5.0 decision tree model to classify loan applicants into five risk categories (1 = lowest risk, 5 = highest risk). The model helps financial institutions make data-driven credit approval decisions.

Key Results:

• Testing Accuracy: 96%
• Correct predictions: 96 out of 100
• Most important feature: Active KPR status (100% importance)

---

Dataset

The dataset contains 1,000+ loan applications with the following variables:

Variable	Type	Description
pendapatan_setahun_juta	Numeric	Annual income (million Rupiah)
kpr_aktif	Factor	Active mortgage status (YA/TIDAK)
durasi_pinjaman_bulan	Numeric	Loan tenure (months)
jumlah_tanggungan	Numeric	Number of dependents
risk_rating	Factor (1-5)	Target variable

Note: The original Excel file is not included in this repository. See data/README.md for data structure details.

---

Model Performance

Confusion Matrix (Testing Set: 100 cases)

Predicted \ Actual	1	2	3	4	5
1	20	0	0	0	0
2	4	13	0	0	0
3	0	0	38	0	0
4	0	0	0	9	0
5	0	0	0	0	16

Performance Metrics

Metric	Value
Accuracy	96%
Correct Predictions	96 / 100
Wrong Predictions	4 / 100
Tree Size	17 rules

Variable Importance

Variable	Importance (%)
kpr_aktif (Active KPR)	100.00
jumlah_tanggungan (Dependents)	75.50
pendapatan_setahun_juta (Annual Income)	68.13
durasi_pinjaman_bulan (Loan Tenure)	57.13

Insight: Active KPR status is the most dominant factor in determining credit risk rating.

---

Decision Tree Rules (Simplified)

If kpr_aktif = YA (Has active mortgage):

• Dependents ≤ 4 → Risk Rating 3 (255 cases)
• Dependents > 4:
  • Income > 248 million → Risk Rating 4
  • Income ≤ 248 million:
    • Loan tenure ≤ 24 months → Risk Rating 4
    • Loan tenure > 24 months → Risk Rating 5

If kpr_aktif = TIDAK (No active mortgage):

• Income ≤ 95 million → Risk Rating 2
• Income > 95 million:
  • Loan tenure > 36 months → Risk Rating 2
  • Loan tenure ≤ 36 months:
    • Income > 201 million → Risk Rating 1
    • Income ≤ 201 million → (further splits, see full output)

Download detailed outputs: Confusion Matrix CSV | Variable Importance CSV | Full Model Summary

---

Predictions on New Applications

Three sample applications were tested:

Application	Income (M)	Active KPR	Tenure (months)	Dependents	Predicted Risk
App 1	200	YA	12	6	4 (High Risk)
App 2	150	TIDAK	64	6	2 (Low Risk)
App 3	300	TIDAK	24	2	1 (Lowest Risk)

---

How to Run the Code

1. Clone this repository
2. Place your Excel file (CreditRisk_R.xlsx) in the data/ folder
3. Run script/model.R in RStudio or R console

---

Connect with Me