Analysis of the dataset UCI Default of Credit Card Clients Dataset, that contains information on default payments, demographic factors, credit data, history of payment, and bill statements of credit card clients in Taiwan from April 2005 to September 2005.
The repository containes three folders:
datasetfolder with the Default of Credit Card Clients dataset;imgsfolder, that containes all the images for the report;resultsfolder, in which all the results are stored.
The script svdd.py containes a implementation of Support Vector Data Description for novelty detection, while the file visualization.py containes some useful tools for visualizing plots and graphs.
Every algorithm used in the machine learning pipeline in the notebook run.ipynb is preceded by a theoretical description.
The analysis is divided into different steps:
- Data exploration: description of the attributes and understanding of each features;
- Data cleaning: processing of data that involves correction of errors, outlier detection, analysis of correlation;
- Outlier detection: identification and removal of rows considered as possible outliers by different algorithm of outlier detection;
- Correlation and dimensionality reduction: analysis of correlated features and application of PCA;
- Class imbalancing managment: application of some tecninques for managing imbalancing in classes;
- Classification algorithms and metrics for evaluation;
- Comments on results achieved.
All the results and the analysis can be found in the file report.html. Nbviewer version available at: Default of credit card clients
Libraries used: Pandas, Sklearn, Numpy, Plotly, Matplotlib, Seaborn, Imblearn.