Effi-FallNet: Harnessing EfficientNet and LSTM for Advanced Video-Based Seizure Detection

Naima Hammad

Abstract: In the rapidly evolving realm of deep learning, the fusion of diverse architectures offers promising avenues for enhancing model performance. This study delves into the amalgamation of EfficientNet and Long Short-Term Memory (LSTM) neural networks tailored for a specialized dataset. While EfficientNet is celebrated for its proficiency in adaptive scaling of convolutional networks, LSTM excels in understanding and retaining long-term dependencies in sequential data. Utilizing a relatively concise dataset, we embarked on this experiment, keen to assess the potential of our unique model combination. Astonishingly, the outcomes exceeded expectations, with our hybrid model showcasing a 100% score across precision, recall, F-measure, and accuracy metrics. Comparative evaluations further cemented our model's dominance, outclassing several state-of-the-art counterparts on the same dataset. This paper provides comprehensive insights into the model's design, execution, and critical evaluation, emphasizing its strengths and potential in real-world applications. However, we also acknowledge the limitations presented by the short dataset, which could introduce risks of overfitting, potentially limiting the model's adaptability to broader contexts. Considering these findings, we project a future where extended datasets and iterative model refinements could set new benchmarks in the field.

Keywords: Fundus images, Deep learning, Cardiovascular diseases, Seagull optimization algorithm, Artificial intelligence