Advancing CO₂ Huff-and-Puff Simulation in Tight Reservoirs: Critical Gaps and Optimization Strategies

Abstract

The rapid evolution of cyber threats and the increasing complexity of network environments have rendered traditional Intrusion Detection Systems (IDS) inadequate for identifying sophisticated attacks. This study investigates the application of machine learning (ML) techniques to enhance the adaptability, accuracy, and efficiency of IDS. Using benchmark cybersecurity datasets KDD Cup 99, NSL-KDD, and CIC-IDS2017 both supervised (Decision Tree, Random Forest, Support Vector Machine, Artificial Neural Network) and unsupervised (K-Means, Autoencoder) models were developed and evaluated under standardized conditions. Model performance was assessed using accuracy, precision, recall, F1-score, and computational efficiency.

Results show that Random Forest and Artificial Neural Network achieved superior performance, exceeding 97% accuracy, while Autoencoders effectively detected unseen attack patterns in unlabeled data. Ensemble optimization further improved accuracy to 98.9% and reduced false positives below 2.5%, demonstrating the robustness of hybrid learning approaches. Integrating ML within IDS substantially enhances threat detection and response capabilities compared with signature-based systems. The findings highlight the practical potential of ML-based IDS for deployment in enterprise and security operations center (SOC) environments, enabling adaptive, intelligent, and scalable cybersecurity frameworks.