Wireless Sensor Networks (WSNs) are used to monitor physical or environmental conditions. Due to energy and bandwidth constraints, wireless sensors are prone to packet loss during communication. To overcome the physical constraints of WSNs, there is an extensive renewed interest in applying data-driven machine learning methods. In this paper, we present a missioncritical surveillance system model for industrial environments. In our proposed system, a decision tree algorithm is installed on a centralized server to predict the wireless channel quality of the wireless sensors. Based on the machine-learning algorithm directives, wireless sensor nodes can proactively adapt their duty cycle to mobility, interference and hidden terminal. Extensive simulation results validate our proposed system. The prediction algorithm shows a classification accuracy exceeding 73%, which allows the duty cycle adaptation algorithm to significantly minimize the delay and energy cost compared to using pure TDMA or CSMA/CA protocols.