International Journal of Science, Engineering and TechnologyAuthors: Aviichal Sharma, Dr. Dolly Sharma
Abstract: We utilized the SPARCS 2015 Inpatient DeIdentified dataset to come up with machine learning models meant to be forecasting total hospital charges and patient disposition outcomes. We made use of both regression and decision tree algorithms for the purpose of the inbound cost prediction. The algorithms we used here are Linear Regression, Ridge, Lasso, SVR, Decision Tree, ElasticNet, KNN, and XGBoost. Besides, in order to do the disposition classification we opted for Logistic Regression, Decision Tree, Random Forest, SVC, Gradient Boosting, KNN, and a Deep Neural Network. We judged model performances with the help of such metrics as R², MSE, accuracy, F1-score, and AUC-ROC. From our research it is seen that XGBoost bested all regression models, registering an R² of 0.9688, while Gradient Boosting secured the highest classification accuracy (87.03%) and the highest F1-score (0.8483). The most substantial determinants that emerged are the length of stay, procedure count, admission type, and DRG codes. Our findings are in line with the finding that the use of machine learning techniques in clinical and operational planning provides still hospital administrators with information on what can be done to optimize costs and what are the most efficient patient discharge strategies. Our results show that data-driven systems have the potential to support value-based healthcare and offer a minimal and scalable model for predictive analytics in hospital settings, as well as provide a scalable blueprint for predictive analytics in hospital settings.
DOI:
