Demand forecasting with artificial neural networks: electric vehicles

This study examines the technical, behavioral, and structural impacts of electric vehicles on insurance systems through a multidimensional approach. It also aims to make a strategic contribution to the industry by developing a demand forecasting model using artificial neural networks. The battery systems, high-voltage electric motors, and software-controlled architecture of electric vehicles create fundamentally different risk profiles compared to traditional vehicles powered by internal combustion engines. These differences necessitate the development of new insurance coverage structures within the sector. Field data indicate that modern risks such as battery failures (42 percent), fires caused by electrical faults (25 percent), and damages linked to charging infrastructure (18 percent) occur frequently. These risks increase the repair costs of electric vehicles by approximately two to three times compared to traditional vehicles, prompting insurance companies to redesign their product structures. Moreover, while 72 percent of electric vehicle users in Türkiye purchase comprehensive automobile insurance, this rate remains at 58 percent among users of vehicles with internal combustion engines, suggesting a cognitive shift in risk perception and coverage preferences. The artificial neural network model developed within this thesis achieved high forecasting accuracy by incorporating numerous independent variables, including income level, environmental awareness, age group, level of insurance literacy, energy costs, and the perceived balance between vehicle price and performance. The model was constructed using the Python programming language and the TensorFlow software library. During model training, the Rectified Linear Unit activation function and the Adaptive Moment Estimation optimization algorithm were applied. The model was trained over 500 iterations, and an early stopping technique was used to minimize the risk of overfitting. The results revealed that in high-income districts of Istanbul such as Kadıköy, Ataşehir, and Maslak, the ownership rate of electric vehicles exceeds 15 percent, while the rate of comprehensive insurance coverage approaches 80 percent. The analysis of consumer behavior showed that younger individuals tend to prefer digital insurance platforms, middle-aged users seek cost-effective solutions, and older consumers prioritize safety-oriented insurance options. Furthermore, environmental concern, perceived social status, and national trust in domestically manufactured brands such as Türkiye's Automobile Initiative Group were identified as key psychosocial factors influencing electric vehicle purchasing decisions. In conclusion, the influence of electric vehicles on the insurance industry is not limited to technical transformations. Rather, it represents a systemic shift that encompasses market dynamics, consumer behavior, and public policy frameworks. It is recommended that insurance companies integrate artificial neural network-based forecasting models into their strategic planning processes, that public authorities increase investment in charging infrastructure, and that manufacturers develop strategies focused on optimizing vehicle price and performance.