This study proposes a deep learning framework for automated gait abnormality detection using CNN-LSTM networks to capture spatial and temporal features. The approach leverages the Gait Abnormality Video Dataset (GAVD) with feature selection and SMOTE to enhance model efficiency and balance classes. Data augmentation improves generalization, and evaluation confirms its effectiveness. This AI-driven method offers a non-invasive solution for clinical diagnostics, rehabilitation monitoring, and assistive technology.