Securing IoT Environments from Botnets: An Advanced Intrusion Detection Framework Using TJO-Based Feature Selection an d Tree Growth Algorithm-Enhanced LSTM

Concerns about cyberattacks are on the rise in the modern digital era, particularly with the expansion of the Internet of Things (IoT). Protecting IoT environments from harmful behaviour requires cybersecurity intrusion detection solutions. The ever-changing nature of infections and the proliferation of attack routes make botnet identification a formidable challenge. Numerous network devices have been targeted by botnet assaults, resulting in significant losses across several industries, due to the fast development of the IoT. Threats posed by botnets to network security are real, and deep learning models have demonstrated promise in efficiently detecting botnet activity in data collected from network traffic. Because of their remarkable capacity to autonomously discover intricate patterns and characteristics inside massive datasets. Unfortunately, the increasingly diverse nature of IoT ecosystems is rendering ineffective the traditional network-level intrusion detection solutions that rely on pre-defined rule sets. A framework to address this issue is presented in this study. When determining if a dataset is suitable for use in transfer learning, our suggested methodology encourages using that dataset as the source domain. The attributes that are most relevant are selected from the pre-processed data using Tom and Jerry Optimiser (TJO). Next, the prediction procedure makes use of Optimised Long-Short Term Memory (LSTM), with Tree Growth Algorithm (TGA) taking LSTM fine-tuning into account. The purpose of this selection procedure is to ascertain whether or not the suggested model is suitable for implementation, providing the optimal course of action in such cases. By selecting an appropriate source domain data set, our evaluation shows that the suggested framework achieves the best accuracy.