The use of machine learning techniques in higher education can be beneficial in optimizing teaching and providing higher institutions with the solutions they need, like monitoring student satisfaction with the instructor's performance. In this study, ten machine learning classification methods are employed on a dataset to predict selected aspects of student satisfaction: Logistic Regression, Linear Discriminant Analysis, Kneighbors, Decision Tree, Naïve Bayesian, Support Vector Machine, Extra Trees, Gradient Boosting, Random Forest, and Multilayer Perceptron. The dataset consists of 5,820 instances obtained from the UCI machine learning repository, and it demonstrates how students rated their instructors in terms of course structure, and behavior. As a result, it was observed that the ten classifiers had better performance in terms of prediction accuracy after balancing the dataset. On the balanced dataset, the ten classifiers were 4% more accurate on average than when they were trained on the imbalanced dataset. In addition, the Extra Trees classifier achieved the highest performance rate based on all the evaluation metrics used in predicting all the targeted features, especially with the balanced dataset. This paper also included the finding of the most important attributes/variables affecting the predictability of the student-satisfaction aspects. As this finding demonstrated, the majority of the important variables were related to instructor characteristics. Moreover, in all cases of the predictions, one variable related to course characteristics (practice-based activities: laboratory work, fieldwork, and group discussions) frequently appeared as the most important attribute compared to other attributes. Thus, and in light of these findings, instructors should plan courses with fieldwork, applications, labs, and group discussion. Instructors should also use up-to-date materials, be well prepared, be friendly, encourage student participation, and give and talk about exam solutions.