Oe-Mdl: Optimized Ensemble Machine And Deep Learning For Fake News Detection

Abstract

The proliferation of fake news in today's digital era has raised concerns about the credibility and trustworthiness of online information. The detection of fake news has become an essential task to protect individuals, organizations, and societies from misinformation and its potential consequences. Existing fake news detection techniques, including rule-based, supervised machine learning, and NLP approaches, have limitations. Rule-based methods lack adaptability, supervised machine learning struggles with generalization, and NLP techniques face challenges in capturing context and nuanced language. To overcome these limitations, this paper introduces the Optimized Ensemble Machine and Deep Learning (OE-MDL) algorithm for effectively detecting fake news. The proposed OE-MDL algorithm addresses the disadvantages of existing techniques by offering several improvements. Firstly, preprocessing methods are employed, including lowercase conversion, tokenization, stop word removal, word stemming, lemmatization, and spell-checking. Additionally, n-grams generation and the computation of term frequency-inverse document frequency (TF-IDF) scores are utilized. By considering a broad range of linguistic and statistical features, OE-MDL aims to capture the nuanced signals that differentiate fake news from real news. Furthermore, OE-MDL combines optimized machine learning (OML) and optimizeddeep learning (ODL) phases for improved classification accuracy and robustness. In the OML phase, base classifiers like optimizedRandomForest, optimizedJ48, optimizedSMO, optimizedNaiveBayes, and optimizedIBk are stacked with anoptimizedMultilayer Perceptron as the Meta classifier. This stacked classifier serves as the base for a bagging classifier, which becomes the classifier for an AdaBoostM1 boosting classifier. In the ODL phase, a Dl4jMlpClassifier serves as the base for a bagging classifier, which becomes the classifier for an AdaBoostM1 boosting classifier. The OML and ODL classifiers are combined using a blending classifier with weighted voting to classify the training set. The trained blending classifier predicts the authenticity of news articles in the testing set. The experimental results demonstrate that the OE-MDL algorithm outperforms existing techniques with the highest accuracy (84.27%), precision (74.17%), recall (85.18%), and F1-Score (79.29%), offering an effective solution to combat the spread of fake news.