Abstract
Fake news is a piece of misleading or forged information that affects society, business, governments, etc., hence is an imperative issue. The solution presented here to detect fake news involves purely using rigorous machine learning approaches in implementing a hybrid of simple yet accurate fake text detection models and fake image detection models to detect fake news. The solution considers the text and images of any news article, extracted using web scraping, where the text segment of a news article is analyzed using an ensemble model of the Naïve Bayes, Random Forest, and Decision Tree classifier, which showed improved results than the individual models. The image segment of a news article is analyzed using only a Convolution Neural Network, which showed optimal accuracy similar to the text model. To better train the text models, data preprocessing and aggregation methods were used to combine various fake-real news datasets to have ample amounts of data. Similarly, the CASIA dataset was used to train the image model, over which Error Level Analysis was performed to detect fake images. model results are represented as confusion matrices and are measured using various performance metrics. Also, to explain predictions from the hybrid model, Explainable Artificial Intelligence is used.
Acknowledgments
This research was supported by the PESU Center for Information Security, Forensics, and Cyber Resilience. We thank the Computer Science and Engineering Department at PES University for their guidance and support. Lastly, we are grateful to our family and friends for their continual encouragement.
Data availability statement
The dataset used for” Automated and Interpretable Fake News Detection With Explainable Artificial Intelligence” comprises text and image-based datasets.
The text-based datasets are available at the following links
The image-based dataset is the CASIA v2 dataset available at