In this century, several academics have been researching on various aspects of human activity detection, recognition, and monitoring. Automatic diagnosis of human physical activity is known as human activity recognition or HAR. Sensor-based HAR technology is among the most significant technologies for monitoring and identifying human activities. Because sensor-based HAR is widely used in regular life and is a fast-expanding field of investigation, it has garnered attention in the computer industry. While electronic gadgets and applications keep on expanding, advancements in artificial intelligence (AI) have enhanced the capability to extract deep hidden data for effective identification and interpretation. However, conventional neural network techniques have trouble distinguishing between similar activities. This article evaluates various deep learning algorithm to perform accurate detection of human activity. In the pre-processing phase, Butterworth filter performs better with PSNR as 61, entropy of 7.92, and correlation coefficient of 0.96 than the other models. Next for segmentation, the Mask R-CNN is better with 1.6 of MSE, 0.9 of MAE, dice coefficient of 0.05 than other segment models. Feature extraction performed using MobileNet-V3 is better for extracting the features and attained NPV of 95.42%. Following that hybrid classifier such as CNN-LSTM performs accurate detection of human activity with the attained accuracy of 97.89%, precision of 94.13%, FDR of 5.87% than the other models. Based on these algorithms, the human activity recognition can be detected accurately than the other algorithms.