Adoption of machine learning algorithm for predicting the length of stay of patients (construction workers) during COVID pandemic

Abstract

The construction sector in a rapidly develop ing country like India is a very unorganized sector. A large number of workers were afected and hospitalized during the pandemic. This situation is costing the sector heavily in sev eral respects. This research study was conducted as part of using machine learning algorithms to improve construction company health and safety policies. LOS (length of stay) is used to predict how long a patient will stay in a hospital. Predicting LOS is very useful not only for hospitals, but also for construction companies to measure resources and reduce costs. Predicting LOS has become an important step in most hospitals before admitting patients. In this post, we used the Medical Information Mart for Intensive Care(MIMIC III) dataset and applied four diferent machine learning algorithms: decision tree classifer, random forest, Artif cial Neural Network (ANN), and logistic regression. First, I performed data pre-processing to clean up the dataset. In the next step, we performed function selection using the Select Best algorithm with an evaluation function of chi2 to per form hot coding. We then performed a split between training and testing and applied a machine learning algorithm. The metric used for comparison was accuracy. After implement ing the algorithms, the accuracy was compared. Random forest was found to perform best at 89%. Afterwards, we performed hyperparameter tuning using a grid search algo rithm on a random forest to obtain higher accuracy. The fnal accuracy is 90%. This kind of research can help improve health security policies by introducing modern computa tional techniques, and can also help optimize resources.