Predicting Depression via Text Analysis: A Design Science Research Approach

Social media platforms generate vast amounts of textual data, offering a potentially rich source for detecting early indicators of mental health conditions like depression, a disorder with significant global impact on well-being. The challenge lies in accurately interpreting subtle linguistic cues within this unstructured data. This study employs a Design Science Research (DSR) methodology to systematically develop, refine, and evaluate machine learning artifacts for predicting depression from text. Specifically, we compare the performance of two prominent models: Support Vector Machine (SVM) and Bidirectional Long Short-Term Memory (BiLSTM). Utilizing a publicly available dataset from Kaggle, the models undergo rigorous text pre-processing, feature extraction, and hyperparameter optimization to maximize their predictive capabilities. Evaluation results demonstrate that the optimized BiLSTM model achieves superior performance across key metrics, including accuracy (97.84 %), precision (99.11 %), and recall (96.52 %), compared to the optimized SVM model (Accuracy: 96.98%, Precision: 98.91%, Recall: 94.96%). These findings highlight the enhanced ability of deep learning models like BiLSTM to capture complex sequential patterns in text relevant to depression. This research underscores the significant potential of integrating carefully designed machine learning techniques into real-world mental health monitoring systems, contributing to innovative, scalable solutions for early depression detection and intervention.