Machine Learning Feature Selection Techniques to Model the Elements of Cash Conversion Cycle and Other Covariates on Hospital Performance

Abstract

This study seeks to contribute by empirically modeling the cash conversion cycle on hospital performance using two supervised machine learning feature selection techniques. In data science, model selection instability is a major concern, especially when dealing with a high number of features. Data mining, such as subset selection technique and regularization (shrinkage) techniques, pays attention to how to extract meaningful information by modeling the raw data. We employed methods such as best subset selection coupled with an exhaustive search using linear regression and shrinkage methods (Lasso, Ridge, and ElasticNet) to model a real dataset. The empirical results indicated that the Lasso outperformed the other shrinkage methods in feature selection even though the average root mean squared error (rmse) was close. Again, Account Receivable Days (ARD), Account Payable Days (APD), Inventory Turnover Days (INV), and Debt Ratio were discovered to be predictors of hospital performance, which are also components of the Cash Conversion Cycle. Finally, the results show that, on average, a day decrease in the hospital’s collection period will decrease performance by 1%, and a one-unit increase in a day in the account payable decrease performance by 0.003 times. Future studies could explore more advanced algorithms, like recursive feature elimination selection methods, to enhance the analysis of CCC on hospital performance. Lastly, it is recommended that hospitals focus on restructuring their cash conversion cycle management, particularly concerning days of account payables.