Many studies have investigated machine learning algorithms to improve the performance of wireless body area networks (WBANs). However, it was difficult to evaluate algorithms in a network simulator because of missing interfaces between the simulators and machine learning libraries. To solve the problem of compatibility, some researchers have attempted to interconnect existing network simulators and artificial intelligence (AI) frameworks. For example, ns3-gym is a simple interface between ns-3 (in C++) and the AI model (in Python) based on message queues and sockets. However, the most essential part is the implementation of an integrated event scheduler, which is left to the user. In this study, we aim to develop a new integrated event scheduler. We present BANSIM, a discrete-event network simulator for WBAN in standard Python that supports deep reinforcement learning (DRL). BANSIM provides an intuitive and simple DRL development environment with basic packet communication and BAN-specific components, such as the human mobility model and on-body channel model. Using BANSIM, users can easily build a WBAN environment, design a DRL-based protocol, and evaluate its performance. We experimentally demonstrated that BANSIM captured a wide range of interactions that occurred in the network. Finally, we verified the completeness and applicability of BANSIM by comparing it with an existing network simulator.