The increasing prevalence of individuals with limited mobility, particularly in ageing populations, necessitates the development of advanced assistive technologies. This study investigates the potential of using the measuring set based on Inertial Measurement Units (IMUs) combined with machine learning algorithms to detect minimal head movements for controlling assistive devices. A series of experiments were conducted using IMU sensors to capture and classify eight distinct head motions based on a registered dataset of approximately 11,000 samples. Various neural network architectures were evaluated, with the 5LSTM+2FC+ATT model achieving the highest accuracy of 85% in motion classification. This research demonstrates that IMUs can effectively facilitate communication between patients with severe disabilities and assistive technologies, potentially enhancing their independence in daily activities. Future work will focus on refining sensor placement and improving model adaptability to individual user needs through personalized motion recognition.