<![CDATA[In recent years, technologies such as big data, cloud computing, and internet networks have grown significantly. Technological advancements are also accompanied by a growth in the number of users of internet-based services, such as cloud services, which grows annually. This growth in user numbers and technological advancements increase the opportunity for cyberattacks through networks, such as theft of user data or information. Therefore, an intrusion detection system is needed as a preventive measure against cyberattacks to protect information on the network. Intrusion detection prevents cyberattacks by using machine learning, which can work effectively in heavy network traffic. Designing an optimal intrusion detection system requires various approaches, such as feature selection and the selected machine learning model. In this study, feature selection was carried out using the filter method, wrapper method, and embedded method. The filter method uses Information Gain (IG) and the wrapper method uses Recursive Feature Elimination (RFE). Then, it was tested with deep learning-based machine learning models such as Deep Neural Network (DNN), Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP), and with traditional machine learning models such as Random Forest (RF) and Logistic Regression (LR). In this research, we contribute by proposing a comprehensive comparative study that evaluates multiple feature selection method and machine learning models to identify the most effective combination for improving intrusion detection system performance. In this study, DNN was able to produce the highest average accuracy of 87.38%. This was followed by MLP, LSTM, and Random Forest with 87.28%, 86.48%, and 86.08%, respectively. Furthermore, the Logistic Regression model had the lowest accuracy value, at 72.34%. Furthermore, the best feature selection method, on average, was the wrapper method, providing a 0.14% improvement compared to the baseline.]]>
