Arabic_Letter_Classification

Exploration into the development of Arabic letter recognition systems using artificial neural networks (ANNs) and hybrid models yielding promising results. Discussion: CNN Achievements: Starting with CNN models, we saw firsthand how deep learning can tackle the task of recognizing Arabic letters, notably improving from our first to our second model. The use of data augmentation was key in helping the models perform well with new data they hadn't seen, addressing initial issues of overfitting. Strengths of Hybrid Models: By blending Random Forest classifiers with features pulled from CNNs, we crafted hybrid models that beat the standalone models in accuracy. This mix of deep learning's insightful feature extraction with the solid classification power of traditional machine learning points to a promising path for better Arabic letter recognition technology. The Complexity Conundrum: Adding more layers to mirror the Arabic alphabet's 28 letters was an insightful experiment but showed that more isn't always better. The performance didn't meet our hopes, emphasizing the need for thoughtful design in our models' architecture. It also showed that that we can use less layers and therefore less computational power and still achieve great results. Link to dataset: https://www.kaggle.com/datasets/mloey1/ahcd1