Self-Organizing Neural Grove for Malware Detection in IoT Edge Devices.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Edge Computing

Background:

• Deep learning models like CNNs excel in classification but require extensive training time.
• Resource-constrained environments necessitate computationally efficient alternatives.
• Multiple Classifier Systems (MCSs) offer faster training and lower overhead compared to deep learning.

Purpose of the Study:

• To introduce the Self-Organizing Neural Grove (SONG), an ensemble learning model.
• To present a novel pruning technique for optimizing classification efficiency in SONG.
• To evaluate SONG's performance on edge computing platforms like the Raspberry Pi 3.

Main Methods:

• Developed the Self-Organizing Neural Grove (SONG) ensemble learning model.
• Implemented a novel pruning technique to enhance classification efficiency.
• Conducted comparative experiments against unpruned MCS, C4.5-based MCS, and k-NN on benchmark and cybersecurity datasets.

Main Results:

• SONG demonstrated superior classification accuracy compared to baseline methods.
• SONG significantly reduced computation time and memory footprint.
• Performance advantages were consistent across diverse datasets and real-world cybersecurity tasks.

Conclusions:

• SONG is a highly efficient and accurate model for edge computing applications.
• The proposed pruning technique effectively optimizes classification performance.
• SONG presents a viable alternative to deep learning in resource-limited settings.