Phishing is an attack method that relies on a user’s insufficient vigilance and understanding of the internet. For example, an attacker creates an online transaction website and tricks users into logging into the fake website to steal their personal information, such as credit card numbers, email addresses, phone numbers, and physical addresses. This paper proposes implementing an extension to prevent phishing for internet users. In particular, this study develops a smart warning feature for the proposed extension using deep learning models. The proposed extension installed in the web browser protects users by checking for, warning about, and preventing untrusted connections. This study evaluated and compared the performance of machine learning models using a malicious uniform resource locator (URL) dataset containing 651,191 data samples. The results of the investigation confirm that the proposed extension using a convolutional neural network (CNN) achieved a high accuracy of 98.4%.

Phishing websites remain a major cybersecurity threat, yet the availability of balanced and feature-rich datasets for evaluating detection models is still limited. While machine learning (ML) and large language models (LLMs) have shown strong potential in URL-based classification, most public datasets provide raw URLs without feature engineering, making reproducibility and fair comparison across models difficult. To address this gap, we present a curated dataset of 111,660 URLs, consisting of 100,000 benign samples (label 0) and 11,660 phishing samples (label 1). Each URL entry is enriched with 22 numerical lexical and structural features (e.g., URL length, domain length, digit ratio, entropy, HTTPS usage). Additionally, three string reference columns (URL, domain, TLD) are preserved for interpretability, and one label column (0 = benign, 1 = phishing), totaling 26 columns. To demonstrate its utility, we evaluate two baseline approaches: a Random Forest (RF) classifier using handcrafted features, and a MiniLM embedding model with Logistic Regression (LR). Both achieved accuracy above 96 % and ROC AUC scores exceeding 0.99 across training, validation, and test splits. This dataset represents an important step toward building reproducible and comparable benchmarks for phishing detection, bridging traditional ML and LLM-based approaches, and supporting future research on adversarial robustness and scalable security models.

Phishing detection systems continue to struggle with real-time responsiveness in distributed web environments. This paper proposes a cloud–edge coordination framework that integrates browser-side Uniform Resource Locator (URL) interception, adaptive cache coordination (ACC), and open ONNX-accelerated inference. The lightweight hybrid model (DistilBERT–MLP fusion) fuses DistilBERT embeddings with standardized numerical features through a multilayer perceptron (MLP). Experiments demonstrate a detection accuracy above 99%, millisecond-level neural inference latency, and end-to-end detection times below 150 ms in live browser environments. ACC reduces cold-cache initialization from multisecond delays to microsecond-level cache lookups. Before ACC, a first-time blacklist fetch takes 7–8 s, and subsequent lookups complete in under 1 μs, with total detection latency ranging between approximately 2 and 110 ms. This behavior demonstrates the effectiveness of ACC in reducing repeated network queries. The experimental results substantiate three principal contributions: an efficient hybrid lexical–numeric detector, an ACC-driven mechanism that mitigates cold-start delays, and a cloud–edge design that strengthens real-time phishing protection at the web scale.

Firewall logs play a vital role in cybersecurity by recording network traffic and flagging potential threats. This study evaluates five machine learning algorithms-decision tree (DT), random forest (RF), extra trees (ET), CatBoost (CB), and AdaBoost (AB)-on a dataset of 65,532 firewall log entries. Models were assessed using accuracy, precision, recall, training/prediction time, and Pearson correlation for feature selection, across multiple train-test splits. The DT model achieved the best performance, reaching 99.45% test accuracy, 97.457% precision, and 93.389% recall at a 7:3 split, along with the fastest training time (0.20642s). We propose real-time flow-level intrusion detection (RT-FLID), novel, lightweight, real-time intrusion detection system that leverages multithreaded processing and flow-level analysis to boost detection speed and scalability. Unlike existing approaches that rely heavily on deep packet inspection or computationally intensive processing, RT-FLID requires minimal resources while maintaining high detection accuracy. The architecture efficiently handles large traffic volumes and dynamically identifies anomalies such as distributed denial-of-service (DDoS) and port scans. Validated on real-world logs, the system maintained high accuracy in critical classes like “deny” and “reset-both.” These findings highlight RT-FLID’s novelty and practical advantages, demonstrating its potential for deployment in high-throughput, low-latency network environments.