Prediction of Patient Visits for Skin Diseases through Enhanced Evolutionary Computation and Ensemble Learning

Skin diseases are an important global public health issue, causing significant health and psychological burdens. Predicting dermatology outpatient visits is essential for optimizing hospital resources and improving diagnosis and treatment methods. Based on machine learning technology and ensemble learning theory, this study integrates four neural network models to construct an optimal prediction model for daily outpatient visits related to skin diseases. To address the issue of local optima entrapment in sand cat swarm optimization (SCSO), an enhanced SCSO is proposed by incorporating the chaotic mapping, the spiral search strategy, and the sparrow warning mechanism. The enhanced SCSO is then utilized to optimize two critical parameters of variational mode decomposition, enabling the extraction of periodic patterns from the skin disease time series. Finally, the enhanced SCSO is applied again to determine the optimal weights for the ensemble model, thereby achieving optimal fusion predictions. We utilized ten years of outpatient data from the dermatology department of a hospital in China, and selected acne, the most prevalent skin condition in the region, as a case study. Experimental results demonstrate that the proposed model effectively combines the strengths of each module, achieving an root mean squared error (RMSE) of 4.43 and an R-squared (R²) of 0.98. Compared to individual models, the RMSE and R² are improved by 79.69% and 36.97%, respectively, effectively overcoming the limitations of single-model approaches. This research provides valuable insights for leveraging medical time series data and optimizing healthcare resource allocation.