Monsif: Automated Arabic Legal Judgment Prediction System for Egyptian Criminal Law

Document Type

Conference Proceeding

Publication Date

2026

Abstract

The Egyptian judicial system grapples with a backlog of nearly 20 million cases, compounded by complex legal documents and limited technological tools for case analysis. This paper introduces Monsif, an automated Arabic legal judgment prediction system tailored for Egyptian criminal law, focusing on drug-related cases. Monsif integrates real-time Optical Character Recognition (OCR) via the ScanDocFlow API, semantic feature extraction using the Gemini 1.5 Flash model, and a fine-tuned XGBoost classifier optimized through grid search algorithm. Monsif employs an XGBoost classifier, fine-tuned with the grid search algorithm to optimize parameters like learning rate, maximum depth, and number of estimators, achieving robust predictions across four judgment categories. Recursive feature elimination selects the top 15 features, ensuring high accuracy and interpretability for legal professionals. Trained on a collected dataset of authentic Egyptian court of cassation rulings, publicly available on Mendeley. The system achieves an accuracy of 81.25% across four judgment categories: Acquittal, Light Imprisonment, Severe Imprisonment, and Life Imprisonment. By providing a web-based platform for automated case analysis, Monsif enhances judicial efficiency and consistency through a scalable and interpretable artificial intelligence system, while also contributing a benchmark dataset and a practical tool for legal professionals.

Link to Full Text

Share

COinS