I recently partook in a datathon organised by the Bristol Data Science society and sponsered by LV=, and with my team of 5 we ended up winning the presentation award and getting the best performance out of all teams.

The task was to create a model to predict the severity of car incidents into fatal and not fatal, with over 75 features and haevily biased data this was no simple binary classification task. My roles involved creating a tableau dashboard (as displayed below) for some visual analysis and the presentation, feature selection for our model and I also tried creating some models.

The performance metric we were to maximise was (ROC + F1 score)/2, and with the binary labels split by 80/20 this was tough for us, however a great challenge.