Enhanced Forecasting Models: Implementation and Comparative Analysis of Sibylla 5.0 with Prophet and NeuralProphet Algorithms

Through a meticulous methodology that includes data collection from historical shipments and predictive scenario testing, followed by a rigorous evaluation of the outcomes, this thesis aspires to contribute valuable insights into the field of predictive analytics. Ultimately, it seeks to demonstrate how sophisticated forecasting models like Sibylla 5.0, when enhanced by powerful tools like Prophet and NeuralProphet, can significantly advance the precision of business forecasting, thereby supporting strategic business decisions and promoting a more dynamic, responsive business environment.

In the rapidly evolving domain of business analytics, the precision of forecasting models plays a pivotal role in driving strategic decision-making and operational efficiency. This thesis delves into the realm of advanced forecasting techniques by focusing on the Sibylla 5.0 model, a cutting-edge predictive analytics tool that harnesses machine learning to blend historical data series with significant business events. The model's primary aim is to refine the accuracy of sales forecasts at both aggregate and individual SKU levels, which is crucial for optimizing inventory management and enhancing customer satisfaction. The research presented in this thesis is twofold. Firstly, it undertakes the task of implementing the Sibylla 5.0 model within a real-world business context, exploring its effectiveness in improving SKU-level forecasting accuracy. The core of Sibylla 5.0's methodology lies in its integration of historical sales data with business realities such as promotions and seasonal trends, using these insights to inform and adjust its predictive algorithms continually. Secondly, this thesis introduces a comparative study of the Sibylla 5.0 model's performance when implemented through two different yet complementary time-series forecasting frameworks: Prophet and NeuralProphet. Prophet, a robust, open-source tool developed by Meta, offers a versatile approach to forecasting with its additive regression model capable of accommodating non-linear trends influenced by holidays and seasonality. NeuralProphet, an extension of the Prophet model, leverages neural networks to enhance predictive capabilities and computational efficiency. By integrating these frameworks, this research aims to not only assess their individual and combined effectiveness in enhancing the predictive accuracy of the Sibylla 5.0 model but also to identify optimal practices and potential challenges in their practical application. The comparative analysis will focus on metrics such as forecast accuracy, computational demands, and ease of integration with existing business systems. Through a meticulous methodology that includes data collection from historical shipments and predictive scenario testing, followed by a rigorous evaluation of the outcomes, this thesis aspires to contribute valuable insights into the field of predictive analytics. Ultimately, it seeks to demonstrate how sophisticated forecasting models like Sibylla 5.0, when enhanced by powerful tools like Prophet and NeuralProphet, can significantly advance the precision of business forecasting, thereby supporting strategic business decisions and promoting a more dynamic, responsive business environment.