Neural Network Method of Analysing Sensor Data to Prevent Illegal Cyberattacks.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Critical infrastructure relies heavily on sensor devices, increasing vulnerability to cyberattacks like data forgery and denial of service.
• Ensuring the security of these sensor networks is a growing challenge due to the rapid expansion of connected devices.

Purpose of the Study:

• To develop an effective method for analyzing sensor data to prevent cyberattacks on critical infrastructure.
• To enhance the security of sensor networks against various malicious activities.

Main Methods:

• Utilized a modified Long Short-Term Memory (LSTM) network for predicting normal sensor data patterns.
• Implemented anomaly detection by analyzing residual values between predicted and actual data.
• Employed a hybrid approach combining predictive modeling with statistical deviation analysis.

Main Results:

• Achieved high accuracy in predicting normal sensor data with an F1 score of 0.90.
• Demonstrated high sensitivity to data changes through anomaly detection via residual analysis.
• Attained a precision of 0.92 for attack detection, with overall accuracy up to 92% and recall up to 89% (AUC = 0.94).

Conclusions:

• The developed method effectively protects critical infrastructure facilities from cyberattacks, even with limited computing resources.
• The hybrid approach offers real-time efficiency and minimal computational costs for robust cybersecurity.
• The method shows significant promise for enhancing the security posture of sensor networks in critical systems.