Today small, battery-operated electrocardiograph devices, known as Ambulatory Event Monitors, are used to monitor the heart's rhythm and activity. These on-body healthcare devices typically require a long battery life and moreover efficient detection algorithms. They need the ability to automatically assess atrial fibrillation (A-Fib) risk, and detect the onset of A-Fib from EKG recordings for further clinical diagnosis and treatment. The focus of this paper is the design of a real-time early detection algorithm cascaded with an A-Fib risk assessment algorithm. We compare accuracy of machine learning schemes such as J48, Naïve Bayes, and Logistic Regression and choose the best algorithm to classify A-Fib from EKG medical data. Though all three algorithms have similar accuracy, the Logistic Regression model is selected for its easy portability to mobile devices. A-Fib risk factor is used to determine a monitoring schedule where the detection algorithm is triggered by the age dependent A-Fib incidence rate inside a circadian prevalence window. The design may provide a great public health benefit by predicting A-Fib risk and detecting A-Fib in order to prevent strokes and heart attacks. It also shows promising results in helping meet the needs for energy efficient real-time A-Fib monitoring, detecting and reporting.