In our everyday life records, human activity identification utilizing MotionNode sensors is becoming more and more prominent. A difficult issue in ubiquitous computing and HCI is providing reliable data on human actions and behaviors. In this study, we put forward a practical methodology for incorporating statistical data into Sequential Minimization Optimization-based random forests. In order to extract useful features, we first prepared a 1-Dimensional Hadamard transform wavelet and a 1-Dimensional Local Binary Pattern-dependent extraction technique. Over two benchmark datasets, the University of Southern California-Human Activities Dataset, and the IM-Sporting Behaviors datasets, we employed sequential minimum optimization together with Random Forest to classify activities. Experimental findings demonstrate that our suggested model may successfully be utilized to identify strong human actions for matters related to efficiency and accuracy, and may challenge with existing cutting-edge approaches.