Intrusion Detection in WSN

Tech Stack

The Problem

• Wireless sensor networks were exposed to stealthy multi-vector attacks evading rule-based detection systems.
• Attack patterns varied drastically across Blackhole, Grayhole, Flooding, and TDMA attack categories.
• High-dimensional sensor telemetry made manual threat identification computationally infeasible at scale.
• Imbalanced datasets caused classifiers to chronically under-detect rare but operationally critical attacks.
• Deployment on resource-constrained sensor nodes required lightweight models without sacrificing accuracy.

Our Solution

• Implemented a full ML pipeline from data preprocessing through model selection and optimisation.
• Applied ANOVA-F and Recursive Feature Elimination to identify the most discriminative sensor features.
• Used SMOTE to balance minority attack class distributions before classifier training.
• Evaluated and systematically compared multiple classifiers, selecting the optimal model per attack type.
• Optimised the final model for deployment on resource-constrained wireless sensor network hardware.

Impact

Achieved high-accuracy attack classification, enabling automated threat response in wireless sensor deployments.