The objective of this project was to determine bad vs. good credit rating with different ML models.
This project was part of the "Hands on Machine Learning" course at Catholic University Eichstätt-Ingolstadt. The objective was to determine bad vs. good credit rating. As the data set stated:
It is worse to class a customer as good (0) when they are bad (1), than it is to class a customer as bad when they are good (1). (FP worse than FN!!) It was important for us to classify the bad credit ratings accuratly.
Our objectives with this project were:
- Create a model that predicts the target value if a new sample has a 'good' or 'bad' credit rating.
- Most important features for credit rating? Reliability.
- Can forward/backward selection improve the performance of the model?
- How well are regression models performing with the categorical data (1hotEncoding)?
We implemented the ML models with ColumnTransformers and Pipelines.
Contains data about people who applied for a loan (e.g.: credit history, purpose of the loan, employment status), and whether the bank deemed them to be a “good” or “bad” credit risk
- instances: 1,000
- features: 20
Data Source from UC Irvine Machine Learning Repository
Please install the environment.yml file as follows:
conda env create -n ENVIRONMENTNAME -f environment.yml
It is important to run all notebooks in following order (to access the variables saved in the Jupyter Notebook kernel):
- Data_Understanding.ipynb
- Data_Preparation.ipynb
- Modelling.ipynb