Application of machine learning for cosmetics sales optimization: analysis and prediction

The thesis deals with the use of machine learning to predict the sales of cosmetic products in the Czech market. The main objective of the thesis was to select and use models for forecasting sales of cosmetic products. The secondary objective was to evaluate the effectiveness of these models. The objectives of the thesis were determined based on the analysis of an interview with a sales manager of a company producing professional hair products. This interview consisted of six main themes aimed at gaining a comprehensive understanding of the company's process in order to identify the company's needs. Real data of the cosmetic company for a three-year period from 2021 to 2023 was used to predict sales. This dataset contained 1 485 921 rows and 15 columns of sales in the Czech market. The machine learning models selected and used include SARIMAX and neural network-based models such as LSTM and RNN. MAE, MAPE and MSE metrics were used to evaluate the performance of the models. The best results were achieved using neural network-based models, especially the RNN model, which showed the highest accuracy based on the metrics. Factors affecting the accuracy of the models include the combination of influencing and non-influencing sales promotions in the same period.