IoT (Internet-of-Things) devices are tremendously widespread in our daily lives and these devices are very often outsourced to third-party companies to save cost. However, it is pointed out that the risk to insert malicious circuitry, called hardware Trojans (HTs), much increases there. The methods using machine learning for detecting HTs at gate-level netlists have been proposed, and those based on ensemble learning models are considered the most effective among them. This paper evaluates the performance of HT detection at gate-level netlists using various machine learning models based on ensemble learning, including random forest, XGBoost, LightGBM, and CatBoost. In particular, we optimize HT features for each machine-learning model and perform HT detection for various gate-level netlists, including intellectual property core netlists. The detailed HT detection results are thoroughly summarized and compared.