Using tree-based models to predict drunk driving at Police preventive checkpoints

Abstract

This Master's Degree Thesis has the following objectives: Understand tree based models and their main advantages and disadvantages Be aware how unbalanced data affects tree based models Deepen drunk driving knowledge in Catalonia Develop a predictive model to detect drunk drivers In this Master Degree Thesis it is described how road safety is a world wide problem and drunk driving an issue to tackle. Original data from Police preventive checkpoints is described and different tree based models are introduced as Classification and Regression Trees, Bagging and Random Forest. It is also detailed the Class Imbalance Problem with several approaches to deal with it. This models are used to develop a predictive model to detect drunk drivers.