Mohammed Helal

I’m a computer science student focused on artificial intelligence and intelligent system design, with hands-on experience building end-to-end deep learning pipelines for real-world applications. I’m driven by the challenge of turning messy data into reliable, scalable models.

In my projects, I applied supervised learning and deep learning to tackle practical problems like highly imbalanced fraud detection, where I built an end-to-end ML pipeline and improved recall using SMOTE and undersampling. I also worked on trip duration prediction by engineering strong temporal and geographic features, then evaluating regression models with strong predictive performance.

My graduation work pushed me into real-world video understanding: I developed a deep learning-based anomaly detection system using PyTorch with dual-stream neural networks combining optical flow and YOLOv5, trained with a Multiple Instance Learning ranking loss for weakly labeled data. I also built a hierarchical deep temporal model for volleyball activity recognition, combining CNN+LSTM person-level embeddings with LSTM team-level classification, packaged with configurable architectures, evaluation, and testing.