Neural Network Modeling and What-if Scenarios: Applications to Various-Term Sales Forecasts  
Author Valentina Kuskova


Co-Author(s) Dmitry Zaytsev; Anna Sokol; Gregory Khvatsky


Abstract We present a complex, theory-driven approach to forecasting automotive market sales with neural networks. We show that market forecast, to be accurate, must be rooted in theory-based interpretation, which allows for an accurate selection of appropriate predictor variables. We have combined several known methods of working with time-series data (missing data imputation, trend adjustment, forecast error decomposition, etc.) with neural network modeling. While the forecasting methods themselves are hardly new, their combination in a single study with solid theoretical is an important innovation in this area of studies, which resulted in an accuracy of over 99% of the forecast against the actual data (as opposed to testing data).


Keywords neural network, predictive models, forecasting
    Article #:  RQD26-122

Proceedings of 26th ISSAT International Conference on Reliability & Quality in Design
Virtual Event

August 5-7, 2021