PENGEMBANGAN MODEL PREDIKSI LAMA RAWAT INAP IBU MELAHIRKAN MENGGUNAKAN PEMBELAJARAN MESIN DI RSUP RATATOTOK BUYAT

• Main Author: Josua Wola, David
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/77869
• Institution: Institut Teknologi Bandung

RB Hospital’s operational indicators show that it consistently experienced increasing resource consumption from 2014 to 2019. The nature of resources is limited, so RB Hospitals need to use them efficiently to optimally serve their purpose in hospitals. Optimizing patients' length of stay (LOS) is one way to optimize hospital resources as it reduces overall LOS, leading to a reduction in resource consumption. This research builds prediction models for post-partum patient LOS, which can be used as decision- support information for patient LOS planning and optimization. The research focuses on baby delivery cases as it is one of the significant cases and resource consumption in RB Hospital. However, not every post-partum patient’s LOS should be optimized, as there is a critical period within the first 24 to 48 hours post-delivery. Thus, it is crucial to identify the "target" patients, which are patients with >2 days of LOS. This research consists of three main parts: research initiation, model building, evaluation, and analysis. The model-building process is further defined in three main steps: data preprocessing, model selection, and model tuning. Data preprocessing is done to ensure that the data used is already in good quality. The model selection process involves building several models used by previous research to identify and select top-performing models. Model tuning is done to further enhance the selected models’ performance by optimizing each hyperparameter and later, the best- performing model is evaluated in the evaluation parts using the test dataset. This research found that the Bayesian hyperparameter tuned-Extreme Gradient Boosting (XGBoost) model performed the best among other models. The best model has an F1 score of 0.804/1.000 with a recall and precision of 0.804/1.000. By using the prediction model, hospitals can identify and cover the planning and optimization of more than 80% of "target" patients by only doing it on 17% of the total patients