" Network \/ TLS Dataset Phishing detection"

"This dataset is collected to support infrastructure-level phishing and malicious hosting detection from encrypted network traffic without decryption. It provides a large-scale collection of network and TLS infrastructure features derived from TLS 1.2 and TLS 1.3 traffic, enabling the detection of phishing URLs directly at the transport layer. The dataset contains 126,063 network sessions and URLs, each represented by 197 numeric attributes describing TLS handshake behavior, protocol versions, server implementation fingerprints, timing statistics, and HTTP\/HTTPS error patterns. Unlike traditional phishing datasets that primarily rely on URL strings or webpage content, this dataset captures low-level transport and application-layer characteristics that reveal anomalous and deceptive infrastructure behavior. The dataset is labeled for binary classification (benign 0 vs. phishing 1). It is suitable for training and evaluating machine learning and deep learning models for network-based phishing detection, malicious hosting identification, and infrastructure anomaly detection. It is derived from TLS traffic traces and supports research on phishing detection in encrypted traffic without payload inspection. Baseline experiments reported in prior studies using Random Forest, XGBoost, and LightGBM achieve accuracies of 97.63%, 97.07%, and 97.40%, respectively, demonstrating the effectiveness of infrastructure-level signals. In addition, experiments with feature optimization and deep learning models (MLP, CNN, LSTM, GRU) show that dimensionality can be reduced by 25\u201350% while further improving detection performance to nearly 98% accuracy with lower computational overhead. Owing to its high dimensionality, heterogeneity, and scale, this dataset is particularly valuable for research on feature selection, representation learning, class imbalance handling, encrypted traffic analysis, and scalable cybersecurity analytics."