Analysis of lost sales quantities with machine learning and time series

This thesis examines major issues such as out-of-stock (OOS) and on-shelf availability (OSA) in the retail sector, using machine learning and time series analyses on data collected over 14 months. It deals in detail with sales and inventory data collected daily from 1779 stores of a large retail chain, as well as daily collected price data from an e-commerce website. Models have been created and analyzed using Linear Regression, Support Vector Machines (SVM), Decision Trees, Random Forest, and Time Series algorithms. The focus of the research is to identify factors affecting lost sales quantities and to use this information to predict future lost sales quantities. This process provides critical data for improving inventory management strategies for retailers. The analyses show that the Lineer Regression model offers higher accuracy and reliability in predicting lost sales compared to other models. It was also found that pricing strategies and competitive pricing have a significant impact on sales volumes. The findings also reveal that the performance of the algorithms drops significantly without sales data. The analysis underscores the importance of the comprehensiveness and accuracy of the data sets used. The results of the study highlight the effects of competitive pricing and effective inventory management strategies on customer satisfaction and sales performance, while showing their potential to create a competitive advantage for businesses in the sector. Future research might include increasing the number of products, incorporating advertising data, and enriching the data with special days. The research provides valuable insights and strategic guidance to industry professionals and academics while guiding on adapting to dynamic market conditions and better meeting consumer needs. It also aims to contribute to increasing operational efficiency of retail businesses and strengthening customer loyalty.