This work focuses on implementing a video-based device capable of detecting normal (standing or sitting) from abnormal (fall state) behaviour using machine learning concepts such as the convolutional neural network (CNN), and long short-term memory network (LSTM). The objective of this work is to achieve an accuracy for all the models in training of at least 90% and above. First, a simple frame differencing algorithm is implemented to extract human motion features. These features are used in the data representation of training and validation subsets. Adding on, models are implemented (individually) using the Google Colab environment to classify the fall from non-fall incidents. This classification is based on the multiple cameras fall dataset (MCFD). Upon training, the evaluation of these models is shown using the LSTM model and the CNN model. These accuracy evaluations of both models are compared to determine the reliability of the system in performing abnormal behaviour classifications for a fall-state detection scenario.