Credit Score Classification Using Machine Learning

R code | Powerpoint

Project Description

The project aims to build and evaluate machine learning models with the goal of predicting an individual's credit score with high accuracy based on their financial history and personal information. Implemented in R, the data was sourced from Kaggle and consisted of a large dataset with 100,000 rows and 28 attributes. Three powerful classification models – Logistic Regression, K-Nearest Neighbors (KNN), and Classification and Regression Trees (CART) – were implemented. Each model was trained with 70% of the data, and their performance was evaluated on the remaining test data, comparing accuracy and sensitivity across different cutoff values.

Project steps:

1. Data cleaning and Preprocessing: The data, initially dirty, required a rigorous cleaning approach. This involved handling of special characters, performing data transformations, addressing outliers, imputing missing values, removing highly correlated data, and combining levels for categorical variables.
2. Logistic Regression: The initial model employed was logistic regression, utilizing forward selection, backward selection, and stepwise methods for variable selection.
3. K-Nearest Neighbors (KNN): Further, KNN classification model was constructed including the variables selected from earlier variable selection methods and determining the optimal k through a systematic search.
4. Classification and Regression Trees (CART): The final model implemented was CART, which included both the minimum error tree and the best-pruned tree.

Conclusion:

The project successfully compared the performance of various models, identifying the superior one. A key takeaway was navigating through unclean data, emphasizing the importance of robust data cleaning processes for accurate model outcomes. The project provided an in-depth introduction to the implementation and evaluation of models in real-world problem-solving scenarios.