A predictive analytics framework using grey-lasso model for hospital readmission / Nor Hamizah Miswan

• Main Author: Nor Hamizah , Miswan
• Format: Thesis
• Published: 2022
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://studentsrepo.um.edu.my/14689/1/Nor_Hamizah.pdf; http://studentsrepo.um.edu.my/14689/2/Nor_Hamizah.pdf; http://studentsrepo.um.edu.my/14689/

Hospital readmission is defined as an admission to a hospital within a certain time frame, typically thirty days, following a previous discharge, either to the same or to a different hospital. Hospital readmission prediction is a challenging task due to the complex relationship between readmission and potential risk factors. At present, previous studies reported modest discrimination ability possibly due to many inherent limitations and complex problem by nature. The selected features to be the input variables for modelling are mainly based on previous models or very shallow exploratory analysis. Moreover, the clinical trials tend to follow established framework when it comes to predictive modelling. Univariate feature selection proceeded by Logistic regression are the well utilized modelling method. On the other hand, most study on readmission prediction has been focus on maximizing the discrimination performance while disregarding the interpretability aspect i.e. actionable insights and reasons that lead to potential readmission. To address the aforementioned limitation, this work tackles the hitherto unexplored problem: hospital readmission framework that have robust preprocessing framework, with reliable and interpretable prediction model. Three approaches were proposed in this work. Firstly, the overall improvement of preprocessing is proposed to enhance the prediction performance. With regards to feature selection in preprocessing phase, Grey-LASSO is proposed with consideration of uncertainty in feature selection by employing Grey relational analysis and LASSO. The features obtained using Grey-LASSO produce minimal features subset to be the input variable in modelling phase. Secondly, machine learning classifiers are used to predict the risk of hospital readmission that have the maximal discrimination performances. Thirdly, interpretable insight based on rule mining is established at the predicted model output to provide certain level of interpretability to be applicable in real clinical setting. The final framework of prediction model, named interpretable rule learning, which harness the elements of interpretability and achieves outstanding performance of faithfulness and human evaluation in terms of, trustworthiness, confidence and understanding. To this end, this work focuses on enhancing the prediction performance as well as provide interpretability insight on the predicted output of modelling.