返回Google圖書搜尋

Simultaneous Imputation and Prediction with High-dimensional Data (SIP-HD)

Zhenzhen JiaJianqiang HuKejia HuQingchen WangNing Zhang

其他書名	A Deep Learning Model for Disease Diagnosis
出版	SSRN, 2022
URL	http://books.google.com.hk/books?id=OKHgzwEACAAJ&hl=&source=gbs_api

註釋Accurate diagnosis directly impacts service quality and resource allocation in healthcare operations. Advanced medical tests can improve doctors' diagnostic accuracy, but they are usually financially costly or infeasible for some patients. To achieve decent diagnostic performance when advanced medical test results are missing, our research proposes a deep learning model, SIP-HD, that simultaneously performs imputation and prediction with high-dimensional data. Compared with traditional two-step models that first impute missing data and then predict disease, our one-step approach directly reduces the imputation and prediction errors accumulated through the two steps. While two-step models treat imputation and prediction independently, in our model, imputation serves the goal of prediction. Specifically, the imputation is part of our deep learning's tuning to achieve the best diagnostic accuracy. Our model is also more advanced than other similar approaches due to its meaningful graph construction, which increases its capacity to handle high-dimensional data and improves model performance. We evaluate our model's performance in comparison with several two-step models, one-step models, and doctors' preliminary diagnoses. In all comparisons, our model stands out in ACC (accuracy), AUC (area under the curve), and F1 score. We recommend our model to partnered hospitals to assist doctors' diagnostic practices, especially in rural areas and telehealth settings where advanced tests are difficult to access.