Credit Risk Evaluator

A machine learning project that compares Logistic Regression and Random Forest classification models to predict loan risk using real-world lending data from 2019–2020.

---

📌 Overview

Credit risk assessment is a foundational challenge in consumer finance. This project trains and evaluates two supervised learning models on a dataset of loan applicants, using 2019 data for training and Q1 2020 data for testing — simulating how a model trained on historical data performs on future, unseen applicants.

---

📊 Results

Model	Data	Accuracy
Logistic Regression	Unscaled	55.87%
Random Forest	Unscaled	63.06%
Logistic Regression	Scaled	69.13%
Random Forest	Scaled	79.64%

Best model: Random Forest Classifier on scaled data — 79.64% accuracy

Key finding: Feature scaling significantly improved both models, with the Random Forest classifier benefiting the most and outperforming Logistic Regression by ~10 percentage points on scaled data.

---

🗂️ Dataset

• Training data: 2019loans.csv — loan records from 2019
• Testing data: 2020Q1loans.csv — loan records from Q1 2020
• Target variable: loan_status (binary classification)

Data sourced from LendingClub loan records. Categorical features were one-hot encoded, and missing dummy columns in the test set were imputed with zeros to ensure consistent feature alignment.

---

🛠️ Tools & Libraries

• Python 3
• pandas, NumPy
• scikit-learn (LogisticRegression, RandomForestClassifier, StandardScaler)
• Jupyter Notebook

---

🔄 Methodology

1. Data Loading — Loaded separate train (2019) and test (2020 Q1) CSVs
2. Preprocessing — One-hot encoded categorical variables; aligned train/test columns
3. Baseline Models — Trained Logistic Regression and Random Forest on unscaled data
4. Feature Scaling — Applied StandardScaler to normalize feature distributions
5. Scaled Models — Retrained both models on scaled data and compared performance

---

📁 File Structure

Credit-Risk-Evaluator/
│
├── credit_risk_evaluator.ipynb   # Main analysis notebook
├── 2019loans.csv                 # Training dataset
├── 2020Q1loans.csv               # Testing dataset
└── README.md

---

💡 Key Takeaways

• Random Forest consistently outperformed Logistic Regression on this dataset
• Scaling is critical for Logistic Regression — unscaled performance was near-baseline
• Training on one year and testing on the next is a realistic evaluation strategy for time-sensitive financial data

---

👤 Author

Randy Crystian Jr
Senior Analyst, Audience Development | Hilton
GitHub