Please refer to Report.pdf for description of project.
Code description:
Only the 'Daily use of a WF credit card' tab (daily-creditcard___detailCategory.csv) of the data file is used. Every customer's sequential purchase history has 4 features: Date (time) Des1 (categorical data) Des2 (categorical data) Payment (numerical data)
Des1, Des2 and Payment can be used as features for the data to build a Recurrent Neural Network model. It is advisable to have larger datasets to better train a deep learning model. Due to small dataset available, we limit ourselves to train only the Des2 feature.
'''ReadData.py Reads sequential card usage data from the 'daily-creditcard___detailCategory.csv' file.
'''rnn.py Contains code for the Recurrent Neural Network model implementation.
Instructions to run the code: (Note: They are specific to LINUX)
- Install python libraries 'numpy' , 'matplotlib' and 'tensorflow' through terminal using following commands: -pip install numpy -pip install matplotlib -pip install tensorflow
(Make sure the above libraries are successfully installed)
-
Run the rnn.py file using command: -python rnn.py
-
Enter masked_id of customer whose purchase you want to predict in the terminal prompt.
Results:
The output is the current, shifted and predicted values for current batch data. Note: The predicts are not very good because the dataset is very small and also it is synthetic data with no purchase patterns.
IMPORTANT We also generate the loss computed for every batch run. The algorithm is successful in the prediction task as the error values generally shows a decreasing trend after every run. (Visible in the plot generated).