Predicting Cyber-Attacks Through the Use of Deep Learning Algorithms

Computer security is a very necessary area due to the new types of cyber threats that continually appear. To try to minimize the risks of cyber attacks as much as possible, it is necessary to use tools and techniques that are capable of detecting them. To do this, a cyberattack detection model is trained, based on deep learning algorithms. The “T5” model has been selected, pre-trained for natural language processing tasks. For the experiments, the “CIC-IDS2017” data set has been used. This dataset has been previously cleaned and adapted to the selected model, and a fine-tuning process has been carried out, from which quite good results have been achieved, exceeding the 97% success rate, for the size. or smallest pre-trained size, “t5-small”, and more than 99% for the second smallest pre-trained size, “t5-base”. Finally, a comparative study is carried out in relation to the evolution of loss rates and success rates, calculated for each size and period. For reasons of simplicity, the SimpleT5 wrapper has been used, which allows the “T5” model to be used with a few lines of code.