Development of a Logistic Regression Model for Fake News Detection in Nigerian Social Media

The rapid growth of the use of social media is posing a significant rate of social, political and economic threats in Nigeria. Communication done online has the nature of being informal, especially in platforms like Instagram, thus complicating the verification of information. It is against this background that this study developed and evaluated a machine learning–based framework for fake news detection on the Instagram social media platform within the Nigerian context. The objectives of the study were to: (i) design a machine learning–based framework for fake news detection on Instagram; (ii) implement the designed framework using Term Frequency–Inverse Document Frequency (TF-IDF) feature extraction and Logistic Regression classification techniques; and (iii) evaluate the performance of the developed model using accuracy, precision, recall, and F1-score metrics. This work adopted an experimental research design. FakeNewsNet repository was utilized to get publicly available benchmark datasets, which contains political and entertainment news - PolitiFact and GossipCop, where data was labeled as fake or real. Nigerian Pidgin English dataset was incorporated into the training process so as to improve contextual relevance and show transfer learning. Under data preprocessing, applied in this research were techniques such as text cleaning, label normalization, and stratified data splitting. TF-IDF, and a Logistic Regression model with class weight balancing was executed under the feature extraction process, where the model was trained and evaluated using an 80:20 train-test split. To simulate Instagram message input and provide instant fake or real news predictions that had confidence scores, a chat model for news verification was implemented. The results showed that the model achieved an overall accuracy of approximately 82%, with satisfactory precision, recall, and F1-score values, indicating effective classification performance. Pidgin English inputs were successfully classified, a key indicator that the model is adaptable to local linguistics patterns. This study concluded that machine learning techniques, when combined with appropriate feature extraction and contextual data, can effectively support Nigerian fake news detection on social media platforms. Recommended in this study is the involvement of larger Nigerian-language datasets, the exploration of advanced deep learning models, and full integration with social media APIs to enhance real-time deployment and enhance the rate of accuracy of detection.