I work on applied projects that connect data, language and systems through real use cases. These include predictive modeling, intelligent interfaces and tools that support analysis, decision-making or automation in practical settings.
đ§ Arabic Sentiment and Dialect Classification
This project focused on building an Arabic-language classification system capable of detecting sentiment, topic and dialect in short social media texts. Using DziriBERT as a base model, I fine-tuned and evaluated it on curated datasets covering various Arabic dialects, including Moroccan, Algerian and Egyptian, along with Modern Standard Arabic. The final model achieved over 90% accuracy in sentiment analysis, outperforming other strong baselines like MARBERT and DarijaBERT.
The system was integrated into a lightweight web interface for real-time predictions. To address dataset imbalance, I extended and balanced existing corpora with additional labeled data, resulting in over 90,000 training samples.
â˝ Soccer Players Face Recognition
A lightweight Flask interface that allows users to upload an image and receive real-time predictions to identify soccer players from labeled image data. The system uses Local Binary Patterns Histograms for feature extraction and a Support Vector Machine for classification. Over 7,000 images were preprocessed and encoded to train and evaluate the model.
𧏠Cancer Mortality Prediction
This project explored the relationship between demographic, socioeconomic and clinical factors and cancer mortality across U.S. counties. Using a dataset of 34 variables, we built a multivariate linear regression model to predict cancer-related deaths per 100,000 residents. The model included features such as incidence rate, poverty percentage, education levels and employment rate.
đ§ââď¸ Legal Case Management Platform
A web application built to assist legal professionals in organising and processing road accident cases. It allows users to manage legal files, schedule hearings, calculate compensation amounts and search for victims, defendants or witnesses, all from a centralised, user-friendly interface.
The platform was developed using Jakarta EE and Spring Boot, with a secure login system, a multi-step case entry form and a responsive UI built with Thymeleaf and Bootstrap. Data is stored in a MySQL database via Hibernate.
đď¸ Android Shopping App
A mobile e-commerce application that allows users to browse products, manage a shopping cart and place orders through a simple, interactive interface. The app features user authentication, real-time product search and a personalised order history per user.
A complete list of my projects can be found on GitHub.