Social media is a platform that accumulates a significant amount of user-generated data without any proper control, which poses a potential threat to individuals and communities. This research paper aims to contribute in three main areas: (1) investigating various techniques used for feature selection in text and image-based data analysis for social media, (2) conducting experiments to demonstrate the impact of different feature selection models on classifier performance for text and images, and (3) developing a novel approach to combine features from text and images for social media data classification. To achieve this, we utilized a dataset consisting of Twitter posts and text-based images from Kaggle. We first employed Optical Character Recognition (OCR) to extract text from images on social media and aligned the images with their corresponding text. We then utilized TF-IDF and chi-square tests to identify the features from the combined image and text data. The experimental results demonstrate that our proposed approach outperforms other techniques and provides an acceptable accuracy rate of up to 89%.

Text Feature selection, Image Feature selection, machine learning algorithm, heterogeneous data, social media data.