Genetic Algorithm-Driven Random Forest Model for Predicting Acute Respiratory Distress Syndrome (ARDS)

Acute Respiratory Distress Syndrome (ARDS) is a critical respiratory condition marked by high rates of mortality, necessitating early and precise prediction to improve patient outcomes. This study introduces a Genetic Algorithm(GA)-Driven Random Forest model to enhance ARDS prediction by optimizing feature selection and hyperparameters. The methodology begins with cohort selection and the definition of clinical and physiological features from an anonymized EHR database. An initial population of Random Forest models, each with a unique feature subset and hyperparameter configuration, is generated. Each model undergoes training and testing, and a fitness score is assigned based on its performance metrics. Using a combination of tournament and roulette selection criteria, the best-performing models are chosen, and genetic operations such as mutation along with crossover are applied to evolve the next generation. This iterative process continues until the stopping criterion is met, and the model boasting peak fitness score is deemed as final selection. The optimized Random Forest model achieved strong performance: an Accuracy of 90.57%, Precision of 90%, Recall of 81.50%, F1 Score of 85.53%, and AUC-ROC of 94.94%. This approach provides a robust tool for continuous ARDS monitoring, enabling earlier diagnosis and timely treatment, potentially reducing mortality rates.