Safa Alsafari

Legal judgment prediction is an emerging application of artificial intelligence in the legal domain, offering significant potential to enhance legal decision support systems. Such systems can improve judicial efficiency, reduce burdens on legal professionals, and assist in early-stage case assessment. This study focused on predicting whether a legal case would be Accepted or Rejected using only the Fact section of court rulings. A key challenge lay in processing long legal documents, which often exceeded the input length limitations of transformer-based models. To address this, we proposed a two-step methodology: first, each document was segmented into sentence-level inputs compatible with AraBERT—a pretrained Arabic transformer model—to generate sentence-level predictions; second, these predictions were aggregated to produce a document-level decision using several methods, including Mean, Max, Confidence-Weighted, and Positional aggregation. We evaluated the approach on a dataset of 19,822 real-world cases collected from the Saudi Arabian Commercial Court. Among all aggregation methods, the Confidence-Weighted method applied to the AraBERT-based classifier achieved the highest performance, with an overall accuracy of 85.62%. The results demonstrated that combining sentence-level modeling with effective aggregation methods provides a scalable and accurate solution for Arabic legal judgment prediction, enabling full-length document processing without truncation.

In this paper, we address the problems of fraud and anomalies in the Bitcoin network. These are common problems in e-banking and online transactions. However, as the financial sector evolves, so do the methods for fraud and anomalies. Moreover, blockchain technology is being introduced as the most secure method integrated into finance. However, along with these advanced technologies, many frauds are also increasing every year. Therefore, we propose a secure fraud detection model based on machine learning and blockchain. There are two machine learning algorithms—XGboost and random forest (RF)—used for transaction classification. The machine learning techniques train the dataset based on the fraudulent and integrated transaction patterns and predict the new incoming transactions. The blockchain technology is integrated with machine learning algorithms to detect fraudulent transactions in the Bitcoin network. In the proposed model, XGboost and random forest (RF) algorithms are used to classify transactions and predict transaction patterns. We also calculate the precision and AUC of the models to measure the accuracy. A security analysis of the proposed smart contract is also performed to show the robustness of our system. In addition, an attacker model is also proposed to protect the proposed system from attacks and vulnerabilities.

Improving Offensive and Hate Speech (OHS) classifiers’ performances requires a large, confidently labeled textual training dataset. Our study devises a semi-supervised classification approach with self-training to leverage the abundant social media content and develop a robust OHS classifier. The classifier is self-trained iteratively using the most confidently predicted labels obtained from an unlabeled Twitter corpus of 5 million tweets. Hence, we produce the largest supervised Arabic OHS dataset. To this end, we first select the best classifier to conduct the semi-supervised learning by assessing multiple heterogeneous pairs of text vectorization algorithms (such as N-Grams, World2Vec Skip-Gram, AraBert and DistilBert) and machine learning algorithms (such as SVM, CNN and BiLSTM). Then, based on the best text classifier, we perform six groups of experiments to demonstrate our approach’s feasibility and efficacy based on several self-training iterations.

Our study explores offensive and hate speech detection for the Arabic language, as previous studies are minimal. Based on two-class, three-class, and six-class Arabic-Twitter datasets, we develop single and ensemble CNN and BiLSTM classifiers that we train with non-contextual (Fasttext-SkipGram) and contextual (Multilingual Bert and AraBert) word-embedding models. For each hate/offensive classification task, we conduct a battery of experiments to evaluate the performance of single and ensemble classifiers on testing datasets. The average-based ensemble approach was found to be the best performing, as it returned F-scores of 91%, 84%, and 80% for two-class, three-class and six-class prediction tasks, respectively. We also perform an error analysis of the best ensemble model for each task.