Problem

The company Ellisphere, a financial institution where i conducted my end of studies internship, was struggling to accurately predict the turnover of French companies. They were using traditional supervised techniques but found that their predictions were not accurate enough to make informed decisions.

Solution

I proposed a hybrid method of supervised and unsupervised techniques to accurately predict company turnover. Our solution involved deploying an ensemble model using k-means, kmedoids, and XGBoost. I used tools like pandas, numpy, scikitlearn, XGBboost for model development and flask for deployment in my implementation.

The unsupervised technique allowed us to identify patterns in unstructured data while the supervised technique helped us capture complex relationships between variables. By combining these two techniques with XGBoost's powerful predictive capabilities, we were able to create a highly accurate model that took into account both structured and unstructured data.

Impact

Increased accuracy of turnover prediction by over 15%
Improved decision-making capabilities for our client

Improving French Company Turnover Predictions