Predict which products Instacart users will reorder using classification algorithms (logistic regression, random forest and gradient boosting)

Project overview:

• Predicted which products each Instacart user will buy in the next order given a history of orders using classification algorithms and engineering time-based features to optimize F1 score
• The ~8 million product orders from Kaggle were stored in postgreSQL
• Trained, evaluated and compared logistic regression, random forest and gradient boosted trees models
• XGBoost was selected as the final model and trained on the full training data on AWS, and performed an F1 of 0.440 on the test set
• Results visualized with Tableau

This repo includes:

• instacart_proprocessing.py: data preprocessing
• instacart_modeling_subsample.ipynb: feature engineering and modeling on subsample of data
• instacart_modeling.ipynb: feature engineering and final modeling on entire training data
• instacart_order_predictions_slides.pdf: pdf of project presentation slides

Note 1: The proprocessing python script is designed to run where postgresql has already been setup and pre-stored with the instacart data (sourced from: kaggle.com/c/instacart-market-basket-analysis/data)
Note 2: Some data files were excluded from the data folder due to github's size limitation