A comparison of time series models for revenue and product demand forecasting in smart fridges.

Hrithik Rai SAXENA

Master's thesis

Abstract:

This thesis offers a comprehensive analysis and implementation of a revenue and product demand forecasting service specifically designed to adapt to the ever-changing environment of Livello GmbH, a cutting-edge startup in the autonomous retail and supply chain sector. This study investigates the necessary alterations that need to be made during the software development stage, encompassing architecture design, data migration, and modeling components. The goal is to decrease response time and efficiently address the unique difficulties associated with time series data generated by smart refrigerators. The accuracy of projections has been reduced due to the unpredictable patterns in revenue time series and the sporadic nature of product demand time series. Additionally, the management of parameters in the current SARIMA model is a cause for worry. To enhance the accuracy of the models used in the forecasting service, various feature engineering strategies are explored. Furthermore, a model retraining strategy has been adopted to preserve computational resources. A thorough methodology will be implemented to evaluate and compare different forecasting algorithms, such as simple exponential smoothing, statistical models, deep learning, and transformer-based methods. This comparison study aims to enhance the advancement of a forecasting system that is both resource-efficient and cost-effective. The research not only improves the durability of Livello's forecasting methodology but also provides valuable business insights that aid in efficient resource allocation. This project serves as a comprehensive guide for Livello stakeholders, offering practical insights and actionable recommendations to inform strategic decision-making. It combines technical advancements with a pragmatic approach. The successful installation of the forecasting system and its subsequent algorithms within the operational context of Livello GmbH provides evidence of their effectiveness and reliability.

Abstract:

This thesis offers a comprehensive analysis and implementation of a revenue and product demand forecasting service specifically designed to adapt to the ever-changing environment of Livello GmbH, a cutting-edge startup in the autonomous retail and supply chain sector. This study investigates the necessary alterations that need to be made during the software development stage, encompassing architecture design, data migration, and modeling components. The goal is to decrease response time and efficiently address the unique difficulties associated with time series data generated by smart refrigerators. The accuracy of projections has been reduced due to the unpredictable patterns in revenue time series and the sporadic nature of product demand time series. Additionally, the management of parameters in the current SARIMA model is a cause for worry. To enhance the accuracy of the models used in the forecasting service, various feature engineering strategies are explored. Furthermore, a model retraining strategy has been adopted to preserve computational resources. A thorough methodology will be implemented to evaluate and compare different forecasting algorithms, such as simple exponential smoothing, statistical models, deep learning, and transformer-based methods. This comparison study aims to enhance the advancement of a forecasting system that is both resource-efficient and cost-effective. The research not only improves the durability of Livello's forecasting methodology but also provides valuable business insights that aid in efficient resource allocation. This project serves as a comprehensive guide for Livello stakeholders, offering practical insights and actionable recommendations to inform strategic decision-making. It combines technical advancements with a pragmatic approach. The successful installation of the forecasting system and its subsequent algorithms within the operational context of Livello GmbH provides evidence of their effectiveness and reliability.

Keywords

revenue forecast, product demand forecast, time series analysis and forecasting, SARIMA, smart fridges

Language used: English

Date on which the thesis was submitted / produced: 9. 2. 2024

Thesis defence

• Supervisor: Markus Eider, M.Sc.

UNIVERSITY OF SOUTH BOHEMIA IN ČESKÉ BUDĚJOVICE

Faculty of Science

Master programme / field:
Artificial Intelligence and Data Science / Artificial Intelligence and Data Science