MLP-SVM: a hybrid approach for improving the performance of the classification model for health-related documents from social media using multi-layer perceptron and support vector machine

In this digital age, due to the rise in medical-related data, there is a concentration in data preprocessing, analysis, prediction, and classification of documents written in natural languages. The aim of this paper is to form a Hybrid model (MLP-SVM) using multilayer perceptron and support vector machine aimed at analyzing and classifying the Afaan Oromo health-related data into predefined labels. Due to its quickness and affordability, automatic classification has gained more interest in this domain. We collected health text data, comments, and posts from the social media in Afaan Oromo. The text preprocessing tasks are applied to data to remove unnecessary texts, punctuation, and numbers. A feature extraction task is applied to obtain a standard Afaan Oromo health dataset using TF-IDF. The demonstration was done using the proposed model, in which the output from the last hidden-layer of the neural network becomes input to the SVM, which finally classifies into various desired classes. Additionally, we employed other methods, such as multi-layer perceptron, logistic regression, SVM, MLP-LR, and the K-NN classifier, for the soundness of the proposed methods in our experiment. Grounded on the comparative study, the proposed model outperformed the other by achieving 98.23% in both accuracy and F1-score. This experimental result presents that the hybrid model is able to enhance the performance of the classifier alone.