Multi-Objective Evolutionary Optimization for Tuning Hyperparameters and Reducing Features in Prediction of Deep Vein Thrombosis

Hyperparameter optimization in machine learning involves finding the best values for hyperparameters that control a learning algorithm's predicting abilities. Exhaustively exploring all possible combinations is infeasible due to the wide range of potential hyperparameter values. Moreover, the number of predicting features in a dataset is often vast, which is common in practical applications like diagnosing or classification in the medical domain. Traditional optimization methods focus on maximizing a single metric, often neglecting trade-offs between multiple performance measures like accuracy and recall. Balancing these metrics is especially crucial in healthcare, where decisions directly impact patients. This work introduces a multi-objective evolutionary approach for simultaneously tuning hyperparameters for the applied machine learning models and reducing feature space to predict Deep Vein Thrombosis, an area with limited exploration. By optimizing accuracy and recall jointly, the proposed approach generates a set of Pareto-optimal solutions from which decision-makers can choose based on preferences or requirements. Experimental results demonstrate that this approach offers Pareto-optimal solutions that outperform those obtained using a complete set of features. Overall, the study showcases the effectiveness of the proposed technique in enhancing predictive modeling for healthcare applications.