Early-stage Malware and Ransomware Forecasting in the Short-Term Future Using Regression-based Neural Network Technique

Abstract

In this study, we propose a predictive model for forecasting future ransomware and malware attacks based on the previous time series data from 2005–2021. We use a time-series regression technique that relies on the neural network algorithm to estimate the forecasting of ransomware and malware attacks in future years over time. Our experiment has applied two hidden layers with the optimal parameter (weight and biases). We modify our model in terms of building time series to predict short-term future values up to 2026. To reach the minimum potential training error, we train our model on 60 epochs to achieve Mean Square Error (MSE) at minimum values. We have achieved the highest accuracy of 99% for forecasting malicious activities (ransomware and malware). The predictive model shows a massive increase in ransomware risk. The current lines of defense cannot keep up with the evolution of ransomware to prevent them.