Optimizing deep learning models with feature engineering: a case study on crash detection with black box data

The availability of large amounts of data enables companies to generate value and services for their customers. This thesis focuses on the application of machine learning methods to data collected by black boxes, devices that collect important data about car crashes. At Generali Italia, an automated decision-making model is employed to analyze real-time multivariate time series data, alerting in case of a crash and providing immediate assistance. The high sensitivity of the model often leads operators to contact customers even in the absence of a real crash. The main objectives of this project are to improve the performance of the model and improve the efficiency of the emergency response process, thereby increasing customer satisfaction and reducing operational costs. The methods used include feature engineering and analysis, identification of correlations and incorporation of improved features into the existing model.