Demand Forecasting Challenge

Retail

Background

Forecasting sales is a standard application of machine learning. There are many approaches to this problem and finding the right one for a given specific task can be tricky.

Our work

We took part in a Kaggle competition to try out various models, including econometric methods, traditional machine learning, and neural networks. As expected, model accuracy highly correlates with its complexity. We also measured how different strategies for feature selection and preprocessing affected the model behavior.

The outcome

At the time of our last submission, we were in the top 20% of Kaggle teams taking part in the contest. What is even more important, we learned a lot about forecasting, and we shared some of this knowledge on our blog.

Technology used
  • Python
  • ScikitLearn
  • TensorFlow
  • Keras