Prediction of Healthcare Demand Using Doctor AI Algorithm

Abstract

Hospitals must estimate the need for healthcare a year in advance and negotiate with insurance providers to establish a budget that will allow them to treat everyone while also ensuring the highest possible income. The hospitals will have to bear the entire cost of treatment on their own if the forecasts prove to be off. The more accurate the prediction, the more compensation is earned. The sequences and labels were created to accurately collect patients’ historical electronic health records (EHRs) on a temporal basis. Deep learning, a subset of machine learning, is employed in this study because of its ability to detect hidden patterns in data that are frequently overlooked by humans. This dissertation attempts to use the Doctor AI algorithm (Choi, Bahadori, Schuetz, Stewart, and Sun (2016a)) to properly estimate future healthcare demand. The results show that for the top 100 categories, the highest accuracy@top- 30 of 94.375% was achieved with 500 nodes in the GRU layers when trained for 20 epochs. On increasing the number of output classes, there was a decrease in accuracy as well. Overall it can be concluded that machine learning can be used for prediction purposes in the future since it can handle EHR data quite efficiently. More research and the comparison between several other methods also need to be done in the future.