Deep Learning Models for Demand Forecasting in Retail.

Achieving high demand forecasting accuracy is one of the important tools for increasing the company's profits in retail. Improving the forecasting process allows to reduce investment in storage and additional transportation, increase the availability of goods on the store shelf and overall customer loyalty, more properly plan the loading of personnel and equipment. In addition, future forecasts can serve as a starting point in shaping strategic actions to change the situation on the market in the long term. The goal of the master's dissertation is to develop a demand forecasting system for Company X, which is one of the leaders on the Russian market for grocery chains. The main part of the work consists of three chapters. The first chapter presents an overview of modern technologies for solving the problem of demand forecasting. The second chapter is devoted to a detailed description of the condition of the task of forecasting demand in Company X, indicating business metrics for assessing the quality of models in terms of accuracy and monetary equivalent, describing existing data along with the stages of their enrichment. The third chapter provides a list of tested architectures of neural networks and the interpretation of the results, also describes the possible architecture of the industrial solution and the entire business process of obtaining a forecast for calculating recommendations for the order to manufacturers of goods.