Feature Engineering for Building Machine Learning Models in Automotive Industry

Feature engineering is crucial for predictive analytics in the automotive industry, where customer behavior is complex and influenced by multiple factors. This paper presents a framework for developing impactful features for vehicle repurchase and service propensity models using cohort-based analysis. By structuring data into observation and performance windows, it establishes clear cause-and-effect relationships. Techniques such as aggregating service histories, dealer interactions, and loyalty patterns extract actionable insights from sales, service, and campaign data. Practical examples, including partial dependence plots, highlight how features like service intervals, dealer proximity, and purchase histories enhance model accuracy and interpretability. The approach captures temporal patterns, optimizing targeting strategies and improving model performance, engagement, and marketing ROI. Future directions include integrating external data, automating feature updates, and real-time deployment Keywords—Feature engineering, Propensity modeling, Machine Learning, Feature creation, Automotive, temporal features, cohort analysis.