A Novel Algorithm to Reduce Machine Learning Efforts in Real-Time Sensor Data Analysis

In the fitness and health fields, wearable sensors generate massive amount of information in big data. The machine learning techniques use the data to assess individuals’ health in real time and identify trends that may lead to better diagnoses and treatments. Applying efficient algorithms to learn from data can aid physicians to evaluate the state of human actions and diagnose the illnesses. The process of discerning valuable information from wearable sensors is a non-trivial task and is an on-going research area. Many research areas have focused on machine learning-based approaches to sensor data for better understanding and meeting people’s needs. However, there are different challenges such as runtime complexity and the number of functions calls associated with these approaches limit us to reach an acceptable accuracy level. To reduce the computational costs of the feature extraction and classification, a novel algorithm is proposed to analyze the variations in the periodic signals. It reduces the learning efforts by detecting any significant changes in the signal. We used the idea of pheromone trail employed in the ant colony optimization algorithm to keep track of the signal updates. The findings of this paper enable the design of a highly effective real-time predictive model for wearable applications.