Enhancing PTSD outcome prediction with ensemble models in disaster contexts

Abstract

Post traumatic stress disorder (PTSD) presents a major mental health issues for individuals who have experienced traumatic events. Timely detection and effective inter vention are vital, as untreated PTSD can result in enduring psychological distress. Accurate identification of PTSD is critical for implementing targeted mental health strategies, particularly among populations affected by disasters. While previous studies have investigated machine learning techniques for PTSD classification, many have encountered challenges related to get better performed model and generalizability. To overcome these limitations, we established a thorough preprocessing pipeline for data cleaning, treating missing values, label encoding, and feature scaling with StandardScaler. The dataset is split into trainin and testing, 80% and 20% accordingly. We created an ensemble model employing a soft voting strategy among various classifiers, including Support Vector Machines, Random Forest, Logistic Regression, XGBoost, LightGBM, and a tailored ANN. This ensemble model achieved an impressive accuracy of 96.76% on a benchmark dataset, significantly surpassing the performance of individual models. The strengths of the proposed method lie in its enhanced robustness through the integration of multiple models, improved generalization across varied data points, and increased accuracy in PTSD detection. This approach provides valuable insights for policymakers and healthcare professionals, utilizing predictive analytics to tackle mental health challenges in vulnerable populations, especially those impacted by disasters.