Securecast: Forecasting CyberAttack using Machine Learning

Targeted cyberattacks have become more sophisticated and frequent in recent years, partly as a result of growing weaknesses in commonly used technology. It is essential to identify and anticipate cyberattacks in advance in order to reduce possible risks and improve network resilience. The exponential growth of the complexity of cyberattacks and digital data have made big data an indispensable instrument for intrusion detection and forecasting. Intrusion detection systems are better able to identify and thwart anomalies and cyberattacks by utilizing unstructured big data. Even though attack prediction has improved, time series and unstructured large data are still underutilized in cyber event forecasting research. The dataset utilized in this study, CSE-CIC-IDS2018, contains a variety of attacks on a real network. In time-series forecasting, methods such as linear regression, long short-term memory (LSTM), and sequential minimum optimization for regression (SMOreg) were employed to build models with ideal parameters. The models' performance was evaluated using a diversity of machine learning techniques, including Support Vector Machine (SVM), Random Forest, and Naive Bayes. Random forest and SVM had the highest success rate, at 90.4%. To evaluate the accuracy of the predicted events, metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were used. The lowest RMSE was shown by linear regression, whereas the lowest MAE was achieved using SMOreg. The goal of this endeavor is to improve detection of cyberthreats and mitigation of security lapses in crucial infrastructure.