Meal Demand Forecasting Using Machine Learning

Abstract

Supply chains are complicated, unpredictably varying systems. Logistics managers are currently confronted with two major challenges: increasingly diverse and fluctuating consumer demand that is difficult to predict. Forecasting demand is critical in the supply chain because it informs essential operational operations such as demand-driven material resource planning (DDMRP), inbound logistics, production, financial planning, and risk assessment. Machine Learning (ML) technologies are increasingly being used for time series forecasting. When forecasting time series data, machine learning is an application of artificial intelligence (AI) that allows forecasting models to learn historical demand patterns and anomalies and enhance future demand prediction accuracy. This reduces the disparity between projected and actual demand levels, giving firms a better sense of what's to come. In this thesis, to emphasize the use of ML methods in demand forecasting, we play the role of logistics service for a meal delivery company which operates in many locations to forecast consumer demand in the upcoming weeks for the raw materials planning, preparing and storing. The proposed method is Multilayer Perceptron (MLP). The suggested approach is compared with other popular time series forecasting approaches, such as Linear Regression, Decision Tree Regressor, Random Forest Regressor, Gradient Tree Boosting, and Multilayer Perceptron. According to the experimental findings, the proposed strategy outperforms the others in terms of performance metrics.