This research article shows a novel technique which used to measure the heart rate (HR) from wearable devices such as fingertip device, wrist type device. For HR monitoring Photoplethysmography (PPG) signal is severely used. HR measurement precision is affected by noise and motion artifacts (MA) at the moment of physical movement. There are many conventional algorithms to decrease the effect of MA and measure the HR variations. In this work, a novel method called multi-model machine learning approach (MMMLA) is used to predict HR. The technique is shown in this work, which primarily trains and tests the algorithm for various features and various data sets. K-means cluster is employed for splitting noisy and non noisy data. This process helps the machine to learn from noisy and non noisy data. After separation, Decision Tree Regression method is employed to fit data and measure HR from test data. In this research feature engineering is completed in different words, a special set of the feature is chosen and check their behavior with the projected model and therefore the error rate for each set of the feature was computed. In this work, model feature is also reduced and check the behavior in estimating HR. For each case root mean square (RMS) error and average absolute error of HR was computed. The minimum average absolute error was traced in this work was 1.18 beats per minute (BPM). The outcome of this work demonstrates that the algorithm has a significant possibility to be used for PPG-based HR monitoring.