Parcel delivery services frequently struggle with failed deliveryattempts due to the absence of customers at the specified addresses. This thesisaims to predict reliable time frames for successful deliveries. For this, machinelearning for improving a relative map to securely encode user location dataand deep learning methodologies for predicting delivery time frames shouldbe implemented and improved to automate and securely determine customerpresence times, thus optimizing the routing efficiency of the logistics service andenhancing delivery success rates. The work is in collaboration with the startupGreen Convenience.