The modeling of health information systems is crucial for enhancing healthcare decision-making, often involving the integration of multiple Internet of Things (IoT) devices. Consequently, robust cybersecurity measures are essential to safeguard information and ensure network security. Even though machine learning-based algorithms have demonstrated their effectiveness and efficiency in various applications, there is a lack of existing works that comprehensively explore the performance of different machine learning algorithms on different classification problems under the same setting. Therefore, this paper explores and compares the performance of four machine learning algorithms, i.e., Random Forest (RF), Naive Bayes (NB), Decision Tree (DT), and XGBoost (XGB), in IoT cyber threat detection for both binary and multiclass classification problems. Experiments conducted on two open-source datasets, NF-ToN IoT and NF-BoT IoT from the Machine Learning-Based NIDS Datasets, demonstrate that the XGBoost algorithm performs best in all scenarios. Further tuning of the XGBoost hyper-parameters significantly enhances its F1-Score performance, achieving up to 98.64% for binary classification and 83.76% for multiclass classification. Among all the tested algorithms, Naive Bayes is the fastest.