Newspapers were the primary source of receiving news. Though they were slow in getting us the news, they were reliable since almost every piece of an article printed in newspapers is proofread. But things are changing rapidly and we are reliant on other sources for news (such as Facebook, Twitter, YouTube, WhatsApp). This paved the way for information, whether it is fake or real, that has never been witnessed in human history before. However, ever since social media boomed and the spread of information became easy, it has been difficult to find and stop the spread of fake and fabricated news. Existing solutions identify fake news usage either or some of the machine learning algorithms. In this work, an ensemble machine learning model is developed using ensemble method and evaluate their performance for the computation time to increase the accuracy of fake news detection using datasets. The experimental evaluation confirms the superior performance of our proposed ensemble learner approach in comparison to individual learners.
Top1. Introduction
Since the advent of social media, the rapid dissemination of information has been paralleled by the spread of misinformation. This rise in misleading content or “fake news” often results from people sharing unverified information or due to their misconceptions. To combat this, many have explored automated systems for detecting fake news using machine learning techniques. Various classifiers, such as Logistic Regression, Multilayer Perceptron, Passive Aggressive Classifier, Decision Tree, Random Forest, and various Naive Bayes classifiers, have been tested for this purpose. Recently, deep learning methods, especially those tailored for natural language processing tasks, have shown promise in offering improved solutions. The study evaluates the efficiency and accuracy of different models in distinguishing between genuine and fake news.
In present days, the hybrid deep learning model built by the authors by combining a Convolutional neural network (CNN) and recurrent neural network (RNN) (Nasir et al., 2021). The primary aim of combining these two models was that they needed good results. Since these are bio-inspired algorithms, they have very good computational power. Since this is a classification problem, the major problem overcome and was bridging the gap created due to a lack of thorough investigation and a lack of combinations of deep learning models built to detect fake news. Then they evaluated the model using two publicly available datasets namely, FA-KES (FakevsSatire) (Brewer et al., 2013) and ISO.
In the logistic regression model, a logistic function for modelling the occurrence of a particular class in the target column is present in the dataset and it is generally of Boolean type. Concerning the proposed solution, it is either true or fake. Even if a problem does not have a Boolean column as an output, LR can be used. This is done by mapping the output to a numerical value, which is between 0 and 1. The basic structure of the Multilayer Perceptron is that it has three layers of perceptron. The first layer is the input layer, the second layer is the hidden layer and the third layer is the output layer. Using a non-linear activation function, each node performs its job except the input nodes. For training, it uses back propagation.
The Passive Aggressive Classifier is one of the models used for applications using large-scale data is a passive aggressive classifier. There are a set of online learning algorithms out of which this is one. The online learning algorithms work with data that is sequential. Decision Tree Classifier belongs to the family of machine learning (Aldwairi, Hasan, & Balbahaith, 2017) models which works by questioning and being answered. The questions are based on the dataset that is used for training the model. The model works this way- firstly it asks for an answer, if it gets the answer, then that path is chosen and further questions of that path are asked. This process continues unless and until a value from the target column is reached. Random forest classifier, one of the algorithms present in the family of machine learning algorithms (Aldwairi, Hasan, & Balbahaith, 2017), itself, is an ensemble of many decision trees that are built as part of its algorithm. The ensemble algorithm is the maximum voting classifier. A theorem proposed by Bayes was based on independent events and this resulted in yet another model called Naïve Bayesian Classifier. Each result produced by the classifiers under this family of machine learning models (Aldwairi, Hasan, & Balbahaith, 2017) is independent of all the results produced by every other model present in the world. Bernoulli Naive Bayes, BernoulliNB implements the naive Bayes training and classification algorithms for data that is distributed according to multivariate Bernoulli distributions; i.e., there may be multiple features but each one is assumed to be a binary-valued (Bernoulli, Boolean) variable. The kaggle dataset is used to implement the proposed system. This paper is organized as follows: Section 2 describes the related work, while Section 3 presents the Proposed Framework of the system. Section 4 depicts the Implementation Results and Evaluation of the proposed system. The conclusion and suggestions for future work are presented in Section 5.