A Sophisticated Framework Designed for Deepfake Detection Using CNN and LSTM for Enhanced Accuracy

Digital literacy is changing very fast, and deepfakes in the context of video call for an important challenge. This study put forward a new approach that combines the strengths of DenseNet v2, LSTM, and MLP architectures to enhance the design of a multi-modal framework for the detection of deepfake videos. We manage an exhaustive survey on deep fake detection that discusses crucial characteristics and techniques needed. This proposed model integrates the benefits of LSTM networks and MLP classifiers with the point birth functionalities of DenseNet v2, thereby forming a robust and efficient framework. The training process involves fine-tuning key parameters so that the model is accurate, and we describe some of the tools we used to preprocess the data and evaluate the models. This work provides a comprehensive performance analysis of the proposed experimental results using precision-recall curves, confusion matrix heatmaps, F1 score comparison bar charts, and delicacy box plots. The proposed multi-modal approach has shown to be better in discovering against the compared models, and thus, is promising for practical usage. This work contributes to both the field of deepfake detection and the wider debate on machine literacy at the intersection with video analysis using CNN algorithm.