HP Supply Chain Optimization

Inspiration

On todays world, were AI and Data Science is taking huge parts of our society, we wanted to introduce ourselfs to the practical uses of Machine Learning. HP's challenge seemed like an interesting context to learn and approach this subject.

What it does

We coded the generation of the Machine Learning Model that can help HP predict the future states of their inventory. This is a Python code that applies the Seasonal Autoregressive Integrated Moving Average with eXogenous regressors (SARIMAX) model to forecast the inventory units of various products in a dataset. It does so by loading and preprocessing the training data, performing Augmented Dickey-Fuller test for stationarity, making the time series stationary, removing outliers using the Z-score method, finding the best SARIMA model parameters, creating a forecast for each product in the test dataset, and generating a final submission file.

How we built it

We started by a simple process of manual data preprocessing, data normalization and model generation to then start using known processes and State-of-the-Art models to try to maximize the desired results. Our model uses the SARIMAX algorithm to predict the inventory units of a given product for a specific week. The model is built using Python and some of its libraries, including pandas, or sci-kit.

Challenges we ran into

Being novices on the topic, we had to learn the structure of a AI program, understand the different data processing steps and learn about different models that adjust to different situations.

Accomplishments that we're proud of

Being novices on the topic, we had to learn the structure of an AI program, understand the different data processing steps and learn about different models that adjust to different situations. We had no knowledge about machine learning models that we could use to face this problem, so we had to do an ample research about different models and test them to find out which of them produced better forecasts.

What's next for HP Supply Chain Optimization

Definitely a handy and useful way to use it. For now we focused on finding the best model and trying to understand why it is the best, but the project needs somewhere to be used. We are thinking to build a WebApp that would let the user keep the data updated and generate predictions for the desired weeks. The model can also be polished but the main focus right now is a front-end.