Predictive Delivery Man Assignment Problem using Deep Learning

Dispatching is a critical part in current online shopping. It relates to how the delivery man assignment should minimize cost along with the service from a source to an end customer with an appropriate scheduled time. The problem arises as neither enough products to deliver nor delivery men are available for dispatch, resulting in suboptimal service and a waste of money. The study aimed to formulate the cost of restaurant dispatching for inducing a deep learning-based solution with the gated recurrent unit recurrent neural network to receive hourly order data and to engage the result for near feature delivery man schedule with minimum cost. The result showed that cost formulation minimized the number of delivery men times the wage per hour with the constraints of each delivery man carrying a maximum of five orders in one way and 11 work hours/day. The deep learning input model used 1078 historical data which were filtered using the Savitzky-Golay method. The root mean square errors of training and testing were 2.35 and 2.41, respectively. Moreover, the number of delivery men every hour was found in a range from one to four people. Furthermore, the deep learning approach saved costs of up to 43.8%.