Predicting the length of stay in a hospital for a particular disease using machine learning algorithms

Abstract

Every single day a large number of patients go to the hospital. But the fact is, facilities available in hospitals are not su cient in comparison with the number of patients. The idea of this paper is to o er a prediction system that will be able to say how many days a patient may stay in a hospital. So that the hospital authority may be able to make a better plan to support a large number of patients. In this project, the focus is to give a statistical overview of the recovery time of di erent diseases in Bangladesh and provide a predictive knowledge based on machine learning algorithms about the possible treatment duration for those diseases. We are hopeful that this prediction system will be a great thing for any hospital authority. They will be able to know the estimated staying duration of a patient in hospital and based on that they will prepare plans to provide support to a larger number of patients. In traditional computing, implementing a system like this is quite impossible because the data here does not follow any algorithmic pattern and that's the reason behind introducing machine learning for this particular task. We tried to accumulate all possible treatments and records of patients and run machine learning algorithms like Linear Regression, Boosted Decision Tree, and Bayesian Regression. We compared the accuracy, mean absolute error and root mean squared error for the results we generated from ML Studio using Linear Regression (LR), Bayesian Regression (BR) and Boosted Decision Tree (BDT) and the results are as follows: Accuracy: LR=0.79, BDT=0.72, BR=0.71 Mean absolute error: LR=0.21, BDT=0.24, BR=0.27 Root mean squared error: LR=0.32, BDT=0.36, BR=0.37