Help to Autocar Insurance Companies: Forecast of Polices Sales on Dynamics of Search Queries

• Published in: International Scientific and Technical Conference "Computer Sciences and Information Technologies" (Online) pp. 564 - 567
• Main Authors: Govorkova, Darya, Boldyreva, Anna, Mogilev, Pavel, Alexandrov, Mikhail, Koshulko, Olexiy
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10.11.2022
• Subjects: advertisement; Approximation algorithms; Companies; Computational modeling; GMDH; Heuristic algorithms; Insurance; insurance companies; Law enforcement; Predictive models; queries
• ISSN: 2766-3639
• DOI: 10.1109/CSIT56902.2022.10000578

There are two traditional ways of quantitative forecast for sales of autocar insurance polices: autoregression on dynamics of sales and regression on dynamics of investments to various channels of advertisement. In both cases the linear models are used. To improve such forecasts, we propose to use: 1) search queries to Internet as a source of independent data reflecting interests of drivers; 2) algorithms of GMDH as a tool, which builds noise-resistant lineal and non-lineal models under limited experimental data. We compare our results with several baselines: Holt-Winters method and ARIMA model in case of autoregression, GMDH models with investments in case of regression. Unlike the typical week and month horizons of forecast we consider month and quarter horizons. In addition, we study possibilities of joint model, where regression includes both investments and queries. Our results show: 1) GMDH-based algorithms without application of investments and queries dynamics decrease errors of forecast (MAPE) approximately in 2.5 times in comparison with Holt-Winters and ARIMA methods; 2) GMDH-based algorithms with application of search queries dynamics decrease errors of forecast (MAPE) in 20% in comparison with application of investments dynamics; 3) joint model with investments and queries doesn't improve forecasts. The proposed technologies may be useful both for autocar insurance companies and for other companies related to sales and advertisements.