## Model Description
This model recognizes static hand signs from the Kurdish Sign Language (KuSL) alphabet, written in the Arabic script used for Central Kurdish (Sorani). It takes a single RGB image of a hand sign and predicts one of 34 character classes.
The model is built on MobileNetV2 (1.0, 224), a lightweight convolutional network designed for fast inference on mobile and edge devices. Starting from the pretrained ImageNet weights, it was fine-tuned for 20 epochs with a learning rate of 5e-5. Its small size makes it practical for real-time applications such as learning apps, accessibility tools, and educational demos.
## Training Data
The model was trained on the **KuSL2023** dataset (Rawf, Abdulrahman & Ali, 2024), which contains 71,400 sign language images across 34 classes. The dataset combines two existing collections:
- **ArSL2018** (Arabic Sign Language): the source of the 23 most commonly used Kurdish letters in the Arabic script.
- **ASL Alphabet**: the source of th…