Detecting Pregnancy Risk Type Using LSTM Algorithm

Abstract

Pregnancy is the most important yet vulnerable phase for all mothers-to-be. Approximately nine months of pregnancy requires special attention from medical workers to monitor the health of the womb. Specifically early detection of risks and diseases that may happen during pregnancy. Risk detection requires understanding, experience, and precise calculations from available dataset. Current methodology of pregnancy risk is manual calculation using KSPR (Poedji Rochyati Score Card). However, manual calculation opens a lot of human error possibilities. Therefore, there is a need to develop a more accurate system using the available data. This study aims to classify the risk of pregnant women using multi-class classification using the LSTM method. Data used in this research are primarily collected dataset from Dinas Kesehatan Kabupaten Boyolali. To create an accurate model, we pre-processed dataset into trainable data for a deep learning model. These processes include balancing data and feature selection. Pre-processed data are then trained and tested. Model hyperparameter are then tuned to provide the best evaluation metric. Final prediction model evaluation metrics collected from the model are 94.63% accuracy, sensitivity 94.57%, precision 94.88%, and F1-Score 94.60%.