Developments in intelligent health care are enabled by various wireless and mobile health applications. The diagnosis and response of remote health data is challenging due to noise, network latency, streaming data, and user privacy. Critical health concerns such as cardiac arrhythmia have potential to induce life-threatening consequences, its effect on quality of life, and its implications for public health systems. Quick identification of arrhythmia aids proper care and reduction of the risk factor. This study focuses on identification of lightweight and accurate machine learning models that can be used for predicting cardiac arrhythmia risks accurately from significant health parameters. We tested Naïve Bayes, K-nearest neighbors (KNNs) and Logistic Regression (LR) models, trained with standard dataset, with 7 features and over 8000 records and tested for its performance. We used multiple significant features to classify the heart risk into three categories: no risk, medium risk and high risk. Naïve Bayes performed poorly with an accuracy of around 61%, while KNN outperformed with an accuracy of about 95% over logistic regression’s 90%.