A Hybrid Deep Learning Model Combining CNN and Extreme Learning Machine for Cyberattack Classification

As ransomware attacks and zero-day exploits grow sophisticated, the need for intelligent, accurate systems to detect such threats becomes clearer. In this paper, a hybrid learning model based on Convolutional Neural Networks (CNNs) and Extreme Learning Machine (ELM) is presented to improve multiclass classification performance for cybersecurity applications. The framework combines CNNs' hierarchical feature learning with ELMs' fast classification. An attention mechanism that assigns weights to each feature based on importance is included in the final model. The hybrid model performed well on the metrics: precision = 0.97, recall = 0.98, and F1- score = 0.97, and, as expected from ELM, inference latency is 15.18 ms/sample. The class weighting method also addresses class imbalance in a generalizable, easy-to-implement way, improving hyperparameters and experimentation for any use case, supporting practicability and sustainability, and enabling wide deployment across resource-constrained environments at low costs. These findings support a scalable paradigm for intelligent intrusion detection systems for the new generation. To this end, experiments on the realistic ransomware dataset UGRansome are presented and designed for benchmarking. The proposed hybrid model is compared against CNN and ELM. The experimental results demonstrate detection accuracy of 97.69% over the baselines tested.