Our team was inspired by StrataScratch's Datathon challenge, offering us the chance to conduct machine learning and data analysis/visualization for operational efficiency on real-world data for companies like Uber.
Our team has developed two distinct models that analyze the most useful factors to predict driver prospects for Uber.
After preprocessing and cleaning up StrataScratch's dataset, we split up into two subteams that worked independently on two different models: one based on a Neural Network trained through technologies and libraries like PyTorch, and one based on a Decision Tree using scikit-learn.
Training two models in the span of 12 hours was not easy, leading our group to face a variety of difficult to resolve challenges. These included deciding best practices in developing our software product using pair programming through services like VSCode liveshare, debugging issues with dependencies and resolving the various issues that come from the rigorous libraries used and time taken to train the models.
Both models performed with impressive performance and accuracy, which were largely due to how successful both our subgroups worked within their respective teams.
Our team learned how to quickly implement and test models, and additionally learned how to deal with larger datasets of data.
As we come to a close with this project, we hope to be able to fit more training data from different companies and different fields to our models, and be able to more accurately and efficiently determine which factors play a prominent role in prospective drivers.