Comparative Analysis of Recurrent Neural Network over Convolutional Neural Network in Predicting R2l Attack in IoT Devices with Improved Accuracy and F1 Score

The main target of this study is to more accurately and efficiently predict the R2L attack on IoT devices using Novel Recurrent Neural Network than Convolutional Neural Network. Materials and Methods: The number of IoT devices that can interact with one another has increased along with the capacity of the internet. This change makes the traditional methods and antiquated data processing techniques for identifying attacks, which are useless. It is challenging to spot attacks on IoT devices and recognise fraudulent activity in its early stages due to the growth in network traffic volume. The study utilized GPower software to calculate sample size requirements. The total number of iterations required for the study was determined using the clincalc tool, which indicated a value of N=20 (10 iterations for each group). For a sample size of 5479 and (ɑ=0.05; power=0.85), each group performed 10 iterations, resulting in a total of 20 iterations for both groups combined. Result: RNN (90.3140%) recognises objects and improves the accuracy over CNN (89.9210%) with significance value P=0.757 (p>0.05), which is statistically insignificant. The F1 score of RNN and CNN are 0.90 and 0.89 respectively. Conclusion: When compared to Convolutional Neural Network (CNN), Novel Recurrent Neural Network (RNN) has more accuracy.