💰 Loan Status Classification (Credit Risk Prediction)

A Machine Learning project to predict credit default risk based on applicant financial profiles.

📌 Background

Financial institutions face significant risks of financial loss due to customers failing to repay loans (default). This project aims to build an intelligent system capable of classifying whether a customer is eligible for a loan (Non-Default/Safe) or poses a risk (Default) based on their demographic and financial history.

📂 Dataset

The dataset used in this project is sourced from Kaggle: Dataset Klasifikasi Status Pinjaman.

• Total Records: ~50,000 rows.
• Target Variable: status_pinjaman (0: Non-Default, 1: Default).
• Key Features: Age, Annual Income, Credit Score, Total Debt, Credit History Length, etc.

🛠️ Methodology & Preprocessing

This project implements an end-to-end data science pipeline:

1. Data Cleaning:

   • Removed the id_pelanggan column (irrelevant unique identifier).
   • Handling Missing Values: Dropped rows containing nulls (percentage was < 5%).
   • Label Encoding: Applied to categorical features such as status_pekerjaan (Job Status), tipe_produk (Product Type), and tujuan_pinjaman (Loan Purpose).

2. Handling Skewed Data (Crucial Step):

   • Identified highly skewed distributions in numerical features (e.g., pendapatan_tahunan, aset_tabungan).
   • Applied Log Transformation (np.log1p) to normalize the data distribution, optimizing it for linear and distance-based algorithms.

3. Data Splitting:

   • Ratio: 80% Train : 10% Validation : 10% Test.
   • Stratification: Used stratify=y to maintain the ratio of the target classes across all splits.

🤖 Modeling

Three Machine Learning algorithms were trained and evaluated:

1. Naive Bayes (GaussianNB) - Used as the baseline model.
2. K-Nearest Neighbors (KNN) - Optimized using GridSearchCV to find the best k and distance metric.
3. Decision Tree - Optimized using GridSearchCV (tuning max_depth, criterion, etc.).

📊 Evaluation Results

The models were evaluated using the Test Set (unseen data). The performance comparison is as follows:

Model	Accuracy	Precision	Recall	Observation
Naive Bayes	72.07%	66.65%	98.61%	Very high Recall, but suffers from high False Positives.
KNN (Tuned)	78.69%	76.61%	88.23%	Moderate and balanced performance.
Decision Tree (Tuned)	88.29%	87.76%	91.48%	🏆 Best Model

Confusion Matrix Analysis (Decision Tree)

The Decision Tree was selected as the final model because it offered the optimal balance between Accuracy and Recall.

• True Positive: Successfully detected the majority of default cases.
• False Negative (Bank Risk): Very minimal (~233 cases out of the total test set), significantly reducing the risk of granting loans to defaulters.

💻 Installation & Usage

1. Clone the repository:

   git clone [https://github.com/faisalsuryasaputra/klasifikasi-status-pinjaman.git](https://github.com/faisalsuryasaputra/klasifikasi-status-pinjaman.git)
   cd klasifikasi-status-pinjaman

2. Install dependencies:

   pip install pandas numpy scikit-learn matplotlib seaborn kagglehub

3. Run the Notebook: Open notebook.ipynb (or your specific filename) using Jupyter Notebook, VS Code, or Google Colab.

👥 Credits

• Faisal Surya Saputra - End-to-End Analysis (Preprocessing, Modeling, & Evaluation)

---

Created as a Final Project for the Artificial Intelligence / Machine Learning Course.