RANDOM FOREST AND MEMORY MODELS IN INTRUSION DETECTION

Abstract

With cyberattacks growing more complex and harder to predict, defending digital networks is becoming a tougher challenge than ever. This study looks at two ways to improve Intrusion Detection Systems (IDS): Random Forest (RF), a popular machine learning method known for being fast and straightforward, and Bidirectional Long Short-Term Memory (Bi-LSTM), a deep learning model designed to analyze sequences of data over time and pick up on subtle patterns. We worked with the CSE-CIC-IDS2018 dataset, carefully prepared to reflect real network traffic, to see how these models perform. Random Forest delivered solid and steady results with 96.8% accuracy, making it a quick and lightweight option perfect for setups where resources are limited. However, it had trouble spotting the more sophisticated, subtle attacks. Bi-LSTM on the other hand did even better by detecting complex intrusions with 98.02% accuracy and strong precision, recall, and F1-scores all close to 97%. Because it processes data both forwards and backwards, it gains a richer understanding of changing threat patterns. What this shows is that while Random Forest works well when speed and efficiency matter most, Bi-LSTM offers the flexibility and strength needed to keep up with evolving cyber threats. Choosing the right IDS means looking beyond just numbers. It is about finding what fits the shifting landscape of cybersecurity. By pairing careful data preparation with advanced AI techniques, this work helps pave the way toward smarter, safer digital networks.