A Novel Deep Learning Technique to Model and Predict the Spread of Cybersecurity Threats

Abstract

Malware, ransomware, and botnet attacks are examples of cyberthreats that are constantly evolving and spreading throughout digital infrastructures, resulting in serious operational and financial harm. The absence of proactive countermeasures frequently leaves organizations unprepared. There is a pressing need to create strong artificial intelligence methods to anticipate cyberthreats because cyberattacks are becoming more frequent and sophisticated. In this investigation, authors present a machine learning-depended (NRM) Non-parametric Regression Model and a deep learning strategy called (DSPM) Deep Sequential Prediction Model to forecast cyber threats across various datasets. Authors use publicly accessible cybersecurity datasets to train and evaluate the suggested models. The suggested models are contrasted with current techniques for predicting cyber threats and assessed using (MAE) Mean Absolute Error. The better prediction ability of the suggested models is illustrated by our experimental findings. While the baseline SVM recorded 25,350 (19%), the PAFE model at (0.75%), the Deep networks model with regularization techniques (Dropout, DropConnect) at (0.43%), Stacked Ensemble, autoencoder, GRU, and MLP (SE-AGM) model at (0.57%), and the model based on gray photos with deep learning (GDMC) at (0.74% and 2.64%), the suggested DSPM and NRM models accomplish MAEs of 1,622 (rate of error 0.62%) and 602 (0.1%), respectively.