Machine learning (ML) for data processing and prediction models has benefited the health sector. One major benefit of training machine learning models is the rapid expansion of data gathering and the conversion of analog data from devices into a numerical value. People working in the technology and medical fields should benefit from this, as it is useful to them. In light of this, we developed a model that forecasts heartbeat rhythms and identifies any irregularities. We employed the prediction models from the first stage to forecast whether heartbeats are regular or irregular on revised datasets created with various data pre-processing methods. Among the twelve ML model deployments, Extreme Gradient Boosting (XGB) produces the best observation with a k-fold mean accuracy (Ma) of 99.14%. In the second stage of the prediction model, the types of heartbeat irregularities are categorized into four classes using the same experimental environment. Also, in the second stage prediction, the mean accuracy of XGB was observed at 98.92 and 99.10% during the test condition. To address the target class data imbalance, consider oversampling techniques like SMOTE and SMOTE-Tomek, respectively. In contrast, performance evaluators’ techniques, like k-fold cross-validation mean accuracy (Ma), Cohen Kappa score (Ck), and recall (Re) strengthen the cause. The low standard deviation (Sd) demonstrates that the model fits exquisitely, with the average deviation being 0.0008 and 0.0002 in both prediction stages, respectively. The model’s ROC-AUC score and the Ck score exemplify the efficiency and robustness of the two-stage prediction model framework.