Elderly and specially-abled individuals can considerably gain from Human Activity Recognition (HAR) systems that have of late recently developed notably owing to the homogenization of the Internet of Things (IoT) and Artificial Intelligence (AI). HAR systems provisioned with several sensors contribute data to homogenization of AI and machine learning (ML) is especially crucial in case of tracking activity levels, such as falling, freezing, stumbling and so on to promote healthy lifestyle. These activities pose significant issues for specially-abled individuals and affect both mobility and security. To address these issues, this study introduces a machine learning method called, Cubic Smooth Spline and Weight Composite Regression (CSS-WCR) Human Activity Recognition for specially-abled individuals to track activity levels. These acts of tracking are still very demanding for the mobility and accessibility of people with special needs. Using pre-processing and feature selection, it is intended to enhance the accuracy and precision of human activity recognition. The proposed CSS-WCR method is split into two processes, namely, pre-processing and feature selection. Initially, sensors are positioned on different parts of the human body to collect the person information. Using Cubic Smooth Spline Data Pre-processing algorithm and Weight Composite Regressive Feature Selection model as the initial features to recognize human activity behavioral patterns among participants are observed. The Cubic Smooth Spline Data Pre-processing algorithm is carried out to remove the noisy data from input sample data points, therefore minimizing processing time and reducing false negative rate extensively. Following which, Weight Composite Regressive Feature Selection is carried out to select the relevant features from pre-processed sample data points with improved precision and accuracy. The usefulness of this study is to improve fitness tracking of the disabled participants by promoting healthy lifestyle and monitoring progress in fitness programs. Besides extending the technical development of HAR and machine learning, this work has inferences for real-life problems of accessibility in monitoring activity levels and physiological data to promote a healthy lifestyle.