Efficient Analysis of Patient Length of Stay in Hospitals Using Classification and Clustering Approaches

Abstract

To maintain resource efficiency and high-quality care, hospitals must estimate a patient’s length of stay (LOS). Machine learning-based prediction algorithms can help. The healthcare industry has become a massive data hub due to its increasing adoption of information technology, but most of this data is kept within the medical institution and not shared with others due to privacy concerns, making it difficult to build effective predictive analytics that need a lot of training data. Using MIMIC database data extraction, we will create two models and offer exploratory and predictive insights. The first model predicts the likelihood of a category event like a patient’s stay. Based on their characteristics and admission circumstances, we will classify patients as short, medium, or long-term stays. The second model groups patients according to the variety of patient-caretaker interactions, such as procedures, drugs prescribed, and inputs taken, which can be used to approximate a patient’s physical or human resource requirements during their stay. We want to blend two models to assess attributes connected to a patient’s background, illness, and therapy. We intend to improve healthcare systems, proactive resource allocation, and patient care by applying predictive analytics and grouping patients by hospital resource utilization. Multinomial Logistic Regression (MLR), Random Forest (RF), and Gradient Boosting Machine (GBM) were developed for the classification problem. Grouping is accomplished using K-Means.