Forecasting Time Series Data

Inspiration

The inspiration for our project was the HP Challenge, which aimed to predict the inventory levels of HP products based on historical sales data. The objective of the challenge was to develop a machine learning model that could accurately forecast future inventory levels, enabling HP to optimize their supply chain management and reduce the risk of stockouts or excess inventory.

What it does

The project is a web application that allows users to predict the inventory levels for a specific week and product. Users input the product number and the year-week combination, and the model generates a prediction of the expected inventory level for that week. The model is trained on historical sales data and uses machine learning algorithms to identify patterns and trends in the data, which are then used to generate predictions for future inventory levels.

The web application provides an easy-to-use interface that enables users to quickly access inventory predictions and make informed decisions about their supply chain management. By providing accurate and reliable inventory predictions, the application can help businesses optimize their inventory levels, reduce costs, and improve their overall efficiency.

How we built it

We used a Flask Python framework to build a server that handles user input features and returns a prediction based on pre-trained ML model; it also renders HTML templates for the web interface. We preprocessed training data on Kaggle with standard techniques, including feature selection, encoding of categorical features and missing values imputation. We created a Dockerfile that will make possible web application deployment.

Challenges we ran into

As we didn't have previous experience with multivariate time series analysis, we realized too late that developing an approach for training a model on data with such a large number of variables (100) and categorical features proved to be impossible in such a short period of time. An additional challenge was related to our team members consistently leaving the team throughout the hackathon, which gradually reduced our efficiency.

Accomplishments that we're proud of

We've learned data preprocessing techniques for time series analysis and ML models for the univariate and multivariate time series forecasting in a short amount of time.