Advancing Fake News Detection: Hybrid Deep Learning With FastText and Explainable AI

Aim:

          To develop a robust and explainable hybrid deep learning framework for detecting fake news by integrating advanced transformer-based models and explainable AI techniques, thereby enhancing classification accuracy, improving model generalization, and fostering transparency in decision-making

Abstract:

         With the rapid spread of fake news across social media and digital platforms, the need for effective detection systems has become increasingly urgent. Misinformation can significantly influence public opinion and decision-making, making it critical to develop reliable tools to identify fake news. This project aims to create a solution for detecting fake news using machine learning and deep learning techniques. By applying natural language processing (NLP) techniques such as stopword removal, tokenization, and lemmatization, we clean and process textual data to improve classification accuracy. We use machine learning algorithms like Random Forest, Extra Tree Classifiers, and Logistic Regression, along with deep learning models like Long Short-Term Memory (LSTM), to build a hybrid system for classifying news articles. The project is divided into two modules: one for classifying the title and one for classifying the text. Additionally, we combine multiple functions into a single pipeline for efficient prediction and apply explainable AI methods, such as LIME and ELI5, to make the model’s predictions more transparent. This approach helps in building a reliable and interpretable system for detecting fake news, contributing to the fight against misinformation online

Proposed System:

         This project introduces a robust fake news detection system combining machine learning, deep learning, and explainable AI techniques. The system preprocesses text using NLP techniques like stopword removal, tokenization, and lemmatization, followed by feature extraction using TF-IDF Vectorization. It is divided into two modules: one for title classification using Random Forest and another for text classification using a hybrid LSTM-GRU model. A unified pipeline integrates these modules for seamless and efficient predictions. Additionally, explainable AI tools such as LIME and ELI5 are employed to provide transparency into the model’s decision-making process, ensuring trust and interpretability. This approach enhances accuracy and reliability, addressing the limitations of existing systems.

Advantages:

• The integration of machine learning, deep learning, and hybrid models ensures high accuracy in detecting fake news.
• Separate modules for title and text classification allow for more focused and efficient processing.
• A unified pipeline streamlines the prediction process, making it efficient and scalable for larger datasets.
• Tools like LIME and ELI5 provide clear insights into the model’s decisions, enhancing trust and usability.
• NLP techniques such as stopword removal, tokenization, and lemmatization improve the quality of input data, leading to better predictions.