Collaborative ambient intelligence-based demand variation prediction model.

Inventory control problem is faced by companies on a daily basis to optimise the supply chain process and for predicting the optimal pricing for the item sales or for providing services. The problem is heavily dependent on a key factor, i.e., demand variations. Inventories must be aligned according to demand variations to avoid overheads or shortages. This work focuses on exploring various machine learning algorithms to solve demand variation problem in real-time. Prediction of demand variations is a complex and non-trivial problem, particularly in the presence of open order. In this work, prediction of demand variation is addressed with the use-cases which are characterised with open orders. This work also presents a novel prediction model which is a hybrid of learning domains as well as domain specific parameters. It exploits the use of Internet of Things (IoT) to extract domain specific knowledge while a reinforcement learning technique is used for predicting the variations in these domain specific parameters which depend on demand variations. The new model is explored and compared with state-of-the-art machine learning algorithms using Grupo Bimbo case study. The results show that new model predicts the demand variations with significantly higher accuracy as compared to other models.