The rapid increase in the deployment of Internet-of-Things (IoT) devices necessitates robust intrusion detection systems. This study evaluates the effectiveness of machine learning models, including Decision Tree, Random Forest and XGBoost in classifying both known and emerging IoT malware threats. Using the CICIoT2023 dataset, we conducted analyses across binary, 8-class and 34-class classifications. Our findings demonstrate that Random Forest consistently outperforms other models, achieving near-perfect accuracy in binary classification and excellent results in multi-class scenarios. Notably, we observed almost no gains in performance beyond a dataset size of 1.5 million records, challenging the notion that larger datasets always equate to better models in malware detection. Furthermore, we investigated machine learning models’ ability to detect zero-day attacks, showing that machine learning techniques offer robust and adaptive solutions for IoT intrusion detection, even in the face of emerging malware attacks.