As cyber threats continue to grow, it is essential to have a strong and adaptable system for detecting malicious activities in network traffic. However, the imbalance between the different types of data in networks poses a significant challenge for traditional detection systems. To address this challenge, our project proposes a new hybrid Intrusion Detection System (IDS) that combines various machine learning algorithms such as XGBoost, LSTM, Mini-VGGNet, and AlexNet. These algorithms are specifically chosen to handle the unbalanced nature of network traffic data. Moreover, we employ the Random Forest Regressor to determine the importance of features in enhancing detection accuracy and interpretability. By identifying the most relevant features in the network traffic data, we can focus our system’s training on these key aspects, improving detection accuracy while keeping computational complexity in check. To tackle the imbalance in the data, we use a combination of oversampling techniques for minority classes and under sampling techniques for majority classes during data preprocessing. This balanced representation of network traffic data ensures that the IDS doesn’t favor common behaviors over rare ones, making it more effective at detecting unusual intrusions. Our research contributes to the ongoing efforts to mitigate cyber threats and protect critical network infrastructures, ultimately enhancing cybersecurity measures in an increasingly digital world.