The threats and injuries caused by falling are significant and pose risks to life among the elderly population. Detecting falls and providing prompt care helps in reducing the risks associated with falling. Technology solutions are being employed for the detection of falls and to enable provisions to access immediate assistance. These solutions require the user to wear specialized equipment continuously that are inconvenient for prolonged usage, especially among the elderly population. To address this concern, numerous research works have focused on developing solutions that are minimally intrusive to the user. Insufficiency of clinical fall dataset limits the investigation of diverse approaches to improve the accuracy using devices that offer user comfort. Algorithms developed through research fall short of extensive datasets to offer reliable detection of falls through a wrist-worn device. However, such algorithms cannot generalize falls using data collected from a specific population. This research discusses an approach for the generalization of falls by the inclusion of kinematic parameters associated with falls. A hybrid detection model comprising of threshold parameters and a machine learning based classifier is proposed. The detection model ignored all the activities of daily living. An accuracy of 92% was observed in the detection of actual falls among non-falling actions with kinematics similar to those of falling action. © 2019 IEEE.