Empowering Mental Health Awareness: Naive Bayes Analysis of Social Media Interactions

This article investigates the efficiency of a Naïve Bayes algorithm for depression identification. We evaluate the model’s performance using a confusion matrix, correctly identifying 1248 members of the “Not Depressed” class, free from false positives, with a precision of 1.00. Although the recall for the “depressed” class is 0.37, it precisely identifies only 54 cases and misclassifies 92 as “not depressed.” This implies that there are rather many false negatives among sad individuals. Still, given its overall accuracy of $93.4 \%$, the model demonstrates remarkable sensitivity to identify non-depressed individuals. Given that the “depressed” class scores significantly lower than the “not depressed” class, the F1-score highlights even more variation. This difference suggests the need for improved feature selection, dataset rebalancing, alternative algorithm investigation, and model tuning to better capture the traits of depressed people. Even though the model performs well at identifying non-depressed cases and increases its accuracy in diagnosing depression, there is still more work to be done, particularly in reducing false negatives. This paper compares the Naive Bayes and Random Forest algorithms for depression diagnosis. Naive Bayes exhibits remarkable accuracy for non-depressed individuals, but it suffers with recall for depressed ones. Random Forest emphasizes the need for dataset rebalancing and model adjustment because it preserves outstanding overall accuracy, delivers improved balance, and boosts recall in depressed conditions.