Predictive Analysis on Credit Score Dataset

This project applies various machine learning techniques to a credit score dataset to predict loan risk levels for customers. By exploring both supervised and unsupervised learning models, the project demonstrates and compares the effectiveness of multiple algorithms, providing a hands-on learning experience.

---

Project Overview

The primary goal of this project was to build, train, and test predictive models on a Kaggle credit score dataset. The dataset includes customer attributes like ID, gender, income, and education as predictor variables. Since a target variable was not provided, I created a new column STATUS with random binary values (0 and 1), representing "low risk" and "high risk" categories. While random assignment does not yield precise real-world predictions, it enables practical model testing and comparisons.

Steps in the Project

1. Data Preprocessing and Cleaning

   • Loaded the dataset and removed columns with minimal impact on prediction, like ID, CODE_GENDER, and FLAG_EMAIL.
   • Converted DAYS_BIRTH and DAYS_EMPLOYED to years and took absolute values to ensure positivity.
   • Transformed categorical columns (e.g., FLAG_OWN_CAR, NAME_INCOME_TYPE) into factors for compatibility.

2. Data Normalization and Encoding

   • Applied normalization to numeric columns for consistent scaling.
   • Encoded categorical columns numerically, converting factor levels to integers to work with all models.

3. Train-Test Split

   • Split data into 70% training and 30% testing sets. This partition provides an unbiased accuracy estimate by evaluating models on unseen data.

4. Modeling

   • K-Nearest Neighbors (KNN): Used KNN with k=10 as a baseline model to predict loan risk, assessing accuracy by comparing predictions against actual labels.
   • Naive Bayes: Implemented Naive Bayes (using the e1071 library), a probabilistic approach that assumes independence among predictors.
   • Decision Tree: Trained a decision tree classifier to predict loan risk, visualized with rpart.plot to observe decision paths.
   • K-Means Clustering (Unsupervised): Applied K-means clustering with two clusters to explore patterns in the data, visualizing cluster centers and groupings.

5. Accuracy Comparisons

   • Calculated and compared accuracies for KNN, Naive Bayes, and Decision Tree models.
   • Created a bar chart with ggplot2 to compare model accuracies visually.

Visualizations

• Decision Tree: Visualized to display classification rules.
• K-Means Clustering Plot: Displays data points grouped by clusters, showing natural groupings.
• Accuracy Comparison Chart: Highlights accuracy for each model to identify the best-performing algorithm.

Libraries and Tools Used

• Data Manipulation: dplyr, lapply, as.factor
• Modeling: class (KNN), e1071 (Naive Bayes), rpart (Decision Tree), kmeans
• Visualization: ggplot2 (bar chart), rpart.plot (decision tree visualization)

Conclusion

This project demonstrates a comprehensive approach to predictive analysis using multiple machine learning models. Although the dataset lacks a genuine target variable, the random target column enables an educational exploration of various classification techniques.