Intelligent Prediction of Network Security Situations based on Deep Reinforcement Learning Algorithm

Abstract

The limitations of traditional network security assessment methods characterized by manual definitions and measurements, data overload, poor performance, and non-negligible drawbacks are addressed in this research. A novel network security system employing a deep learning algorithm is proposed to overcome these challenges. The research unfolds in three key phases. First, a deep self-encoding model is developed to distinguish various network attacks effectively. Subsequently, the creation of missing measurement weights enhances pattern detection, even when dealing with a limited number of training samples. Finally, the model assesses and computes attack issues, assigns impact scores to each attack, and determines the overall network security value. Experimental results demonstrate that the deep auto encoder-based deep neural network (DAEDNN), in conjunction with the proposed unique oversampling weighting (UOSW) algorithm, significantly outperforms traditional methods such as decision trees (DT), support vector machines (SVM), and long short-term memory (LSTM) models. The F1 score of UOSW surpasses these models by approximately 2.77, 10.5, and 5.2, respectively. The deep self-encoding model employed in the proposed system offers superior accuracy and recall rates, leading to more precise and efficient measurement results.