Predicting the amount of medical waste using kernel-based SVM and deep learning methods for a private hospital in Turkey

Abstract

Successful medical waste management requires accurate forecasting of the amount of waste generation. In the case of increasing the number of independent variables, traditional regression methods are insufficient to predict the amount of waste production. On the other hand, methods such as Kernel-based Support Vector Machine (SVM) and Deep Learning, which have more complex algorithms, give more successful results in predicting the amount of medical waste. In this study, the amount of medical waste for a private hospital in Antalya, Turkey, was predicted using Kernel-based SVM and Deep Learning methods. Epanechnikov function for Kernel-based SVM and Maxout activation function for Deep Learning method were used. The number of surgeries, number of outpatients, number of inpatients, number of intensive care patients and number of intensive care days were determined as the model inputs. In addition, the proposed models' performance was evaluated based on Root Mean Error (RMSE), Mean of Absolute Error (MAE) and R-squared. Although both algorithm results are successful in terms of evaluation criteria, it is seen that Deep Learning method (RMSE = 0.094, MAE = 0.079 and R2 = 0.466) is more successful than Kernel-based SVM (RMSE = 0.264, MAE = 0.202 and R2 = 0.221). It is thought that Kernel-based SVM and Deep Learning algorithms can successfully interpret the relationship between the amount of medical waste production and model inputs and play an efficient role in the planning of medical waste management.