Performance Evaluation of Naïve Bayes and SVM for Detecting Mental Health and Depression Sentiment in Social Media X

Abstract

Highly effective strategies for early detection of mental health issues are urgently needed, given the increasing incidence of mental health issues among adolescents. This study investigates sentiment analysis as a tool for identifying depression in social media accounts. Selected reviews about depression will be used to compare two machine learning classifiers: Naive Bayes and Support Vector Machines (SVM). Data will be processed and vectorized using the TF-IDF technique, divided into 70% training and 30% testing. Algorithm performance will be evaluated using Recall, accuracy, precision, and F1 score. Support Vector Machines significantly outperformed the Naive Bayes model. SVM achieved 92.93% accuracy compared to 79.01% accuracy, 93.00% precision, and 92.93% recall. For Naive Bayes, the precision and recall were 80.71% and 80.71%. The determining factor for the superior machine learning performance of SVM is its ability to detect complex and non-linear linguistic aspects present in expressions of depression and mental health. These results demonstrate how advanced machine learning algorithms such as Support Vector Machines (SVMs) can form the basis for proactive and scalable mental health support systems.