Rumor Prediction System

The rapid proliferation of multimodal misinformation across social media platforms presents a significant challenge to information integrity and social stability [9].To address this threat, this project proposes a comprehensive detection system integrating dedicated models for image, video, and audio analysis [13]. The image model utilizes a ResNet18 convolutional neural network (CNN) implemented in PyTorch, leveraging transfer learning and Grad-CAM to provide explainable AI visualizations that highlight the most influential regions for a classification deci- sion [9], For video analysis, the system employs a frame-based ap- proach using OpenCV for extraction and key frame sampling, en- abling suspicious frame detection to pinpoint specific instances of tampering within a video stream [13], The audio model processes content using the Librosa library to extract Mel Spectrograms and Mel Frequency Cepstral Coefficients (MFCC), training a CNN enhanced by data augmentation techniques—such as noise addition and pitch shifting—to ensure robustness against diverse acoustic environments [13], By aggregating predictions across these modalities, the system provides veracity labels accompanied by confidence scores and visual heatmaps, significantly enhancing both the accuracy and transparency of fake news detection in complex multimedia environments [13].