Comparative Analysis of Machine Learning and Deep Learning Techniques for Social Media-Based Stress Detection

With the rapid replacement of spoken communication by text-based modes, recent developments in digital technology have enabled analytical research for stress detection based on social media textual inputs. Nevertheless, there is a dearth of publicly accessible benchmarks. By combining two top-notch datasets derived from Reddit and Twitter, were able to accomplish our aim of stress detection using social media texts. There are three primary benefits: 1) describe in detail collected, preprocessed, and extracted features from each dataset; 2) compare the results of various DL (Deep Learning) and (Machine Learning) ML-based methods namely SVM (Support Vector Machine), CatBoost (Categorical Boosting), RNN (Recurrent Neural Network), RF (Random Forest), BNB (Bernoulli Naive Bayes), LR (Logistic Regression), CNN (Convolutional Neural Network) and DT (Decision Tree) using these datasets as a new standard in a field; 3) evaluate performance of these models in terms of F1-score, recall, accuracy, and precision. The ML-based decision tree model gets 95% accuracy that higher than other models. Models based on ML perform better than Deep-based techniques, according to extensive trials.