Repository for final project in Predictive Analytics.
I compete in Kaggle’s “Predict Future Sales” data competition, where the goal is to predict 1C Company’s sales for November 2015 using data from January 2013 to October 2015. The key challenge of this competition is that the predictions must be by store and product, requiring a total of 214,200 distinct predictions. I employ a variety of statistical techniques to approach this forecasting problem, comparing their strengths and weaknesses. I find that machine learning methods (NeuralNet and LightGBM) outperform traditional forecasting methods in this prediction problem.
For replication purposes, I used R version 4.4.2 and Python version 3.12.10. The requirements.txt file lists the Python packages needed to run the code. I recommend creating a virtual environment for installing these packages. Be warned that the machine learning code can take several hours to run.