Human activity recognition (HAR) research has become increasingly important to healthcare systems recently. The HAR's accurate activity classification results have wide-ranging applications that improve the performance of the healthcare system. The system forecasts abnormal actions based on user movements, and the HAR findings are helpful in keeping track of a person's health. The abnormal activity forecasts made by the HAR system improve healthcare monitoring and decrease medical issues among users. To achieve the best possible result, multiple machine learning and deep learning models have been developed for this area. As more effective and powerful deep learning approaches are developed, the employment of older, more conventional machine learning algorithms has declined in popularity. Convolution neural networks (CNN), long short-term memory (LSTM), recurrent neural networks (RNN), and other deep learning methods are frequently used. On numerous public datasets, the developed algorithm has been evaluated to show its efficacy, and in certain cases, it has outperformed stateof-the-art performance. Some tools are presented and discussed to ensure the suggested algorithm's effectiveness for continuous real-time human action recognition and to verify the algorithm's variability. In this article we will use hybrid model to improve performance of deep convolution neural networks (ConvNets) to detect human activities.