Current Steps

[JoeyChor]

The points below are the things we should complete before the final project is due. This doesn't include anything that we should all contribute, like expanding on the report and creating the presentation.

• Data visualization
  • Generate even more graphs (preferably varied)
  • Understand the data better
• Feature engineering
  • (soft depends on data visualization as we need to understand the data better to extract more features)
  • Extract/derive even more features
  • Try to be creative and not be bound by existing data here (check the table in the README)
• Feature ranking
  • (soft depends on feature engineering as we need to the extracted features for them to be ranked)
  • Perform more Shapley analysis, maybe with different explainers
  • Clean up the PCA and LDA code for feature ranking
  • Redo PCA and LDA for new engineered features
  • Maybe generate a graph for PCA that shows optimal number of features
• Prediction model
  • (soft depends on feature engineering for new features and feature ranking for potential correlated features)
  • Continue iterating through multiple combinations of features
  • Redo on new engineered features
  • Clean up the Bayes and kNN models
  • Add Parzen windows and neural network models
  • Apply best practices from her slide
• Model evaluation
  • Generate metrics Rahul suggests: accuracy, precision, recall, F1 score
  • Generate graphs ROC curve, confusion matrix, and others
  • Interpret the results
  • Strength, weaknesses of models
  • Recommendation to solve business problem (see Rahul's notes)

[satyam04sharma]

We need to predict the probability that these new buyers would purchase items from the same merchants again within 6 months and write in words for conclusion section