Changes during the life cycle of IoT projects may occur, the security of an Internet of Things (IOT)-based system might be compromised due to a lack of awareness and the incorrect selection of relevant test cases. As a result, there is an increasing need for a solution that categorizes security requirements and ensures the efficacy of continuous security testing on IoT devices. Deep learning has been applied successfully in a variety of applications, including requirement classification automation and IOT systems. This study contains two major components. First, we build a dataset. Second, we used Long Short-Term Memory (LSTM) to classify the security requirements of IOT devices into multi-classes as defined by ISO 25010: confidentiality, integrity, non-repudiation, accountability, and authenticity. The model was trained and evaluated using our collected dataset. The testing findings indicated that the suggested classification model can achieve a classification performance of 79% accuracy.