Reto JDA (#43)

We've trained multiple Machine Learning models to predict the sales quantity for each point of sale that can be queried to our API. In particular, the ones for the three months following training data

Comment

Inspiration

We heard about the Machine Learning challenge by JDA at the beginning of the Hackathon. Two of our team members had some previous experience with the technologies that could potentially solve the problem and we decided to try it.

What it does

We predict the sales quantity for a point of sale by calling our API. We prepared for different scenarios on incomplete information and outlier data. Additionally, we've populated the predicted sales quantity values for the test data.

How we built it

We cleaned the data for training by adding and removing features, depending on how we wanted to work with them and their relevance. The data is also scaled so that regression models perform better. Then the model was trained and was ready to predict on new data.

We've also created a front-end that allows the client to provide values for a single point of sale and query the model. The API was built with Flask and the UI with JavaScript, HTML and CSS. The data is processed with pandas and models are trained with Scikit-Learn

Challenges we ran into

These are some of the challenges we faced:

  • Work with incomplete tuples and outliers for training the model.
  • Choose the right model among the options we considered.
  • Optimize the chosen model by varying its parameters.

Accomplishments that we're proud of

We trained and were able to predict on new data.

What we learned

We learned how to use Scikit-Learn and web technologies.

What's next for Reto JDA (#43)

Following steps would be to improve the model that we trained for better performance in the future.

Share this project:

Updates