Hospital-Readmission-based-on-DiabetesData

1. Problem Statement Diabetes is a widespread chronic disease that is accompanied with irregularities of blood glucose levels due to problems related to insulin. For diabetes; the cost analysis estimates that $250 million can be saved across 98,000 diabetic patients by incorporating predictive modeling and prompting greater attention to those who were predicted to get readmitted. Hospital readmission rates are an indicator of hospital quality as well as a means to reduce healthcare costs. High readmission rate is an indication of poor hospital quality, poor treatment and results in the hospital getting penalized. Hence by identifying the factors that lead to higher readmission and being able to predict if a patient is going to be readmitted, the treatment provided by the hospital can be changed, to avoid a readmission and thereby the quality of healthcare provided to the patient can be largely improved, as well as billions of dollars can be saved.
2. Data Science Methodology By leveraging the dataset, we will perform extensive Exploratory Data Analysis and perform feature engineering to identify the suitable attributes that are having significant effects on the target variable i.e readmission. After doing exploratory data analysis (EDA) and feature selection, we can build Classification and Clustering Machine Learning models in order to predict whether the patient is going to be readmitted or not. We are planning on implementing Classification algorithms like Logistic Regression, SVM, Feedforward Neural network, etc. and clustering techniques like Kmeans and Hierarchical Agglomerative Clustering algorithms in order to produce the optimum number of clusters. We will then go ahead with the clustering algorithm that produces clusters with the highest purity and least entropy. Extensive model fine-tuning can be done in order to build a model with the best accuracy and recall. The recall is more important in this case as it is very important to minimize Recall by reducing the False Negatives, as incorrectly classifying a patient as ‘Not Readmitted’ when he/she is going to be readmitted can be very problematic and is highly undesirable.
3. Project Objective The objective of the project is to help the hospitals accurately predict if a patient is going to get readmitted after discharge. The predictions made will help the hospital make informed decisions on the necessary treatment process alterations in order to reduce patient readmission rate. In this context, you can see different objective functions for the problem. We have to try to figure out situations where the patient will not be readmitted, or if they are going to be readmitted in less than 30 days. Make your choice and let's help them create new approaches for the problem. We will try to predict which factors are more important in determining if a patient might get readmitted within 30 days.
4. Expected Results By accurately predicting if a patient is going to be readmitted, the hospital will be able to (taking into account the patient demographics, the treatment procedure and medicines) make necessary treatment changes to avoid the patient readmission. This reduction in patient readmission rate will help the hospital save billions of dollars and also improve their customer rating. The process of EDA followed by model building, fine tuning and performance evaluation is an iterative process which can be carried out until the best and desirable results are achieved.
5. Dataset Source Prediction on Hospital Readmission Classification using Diabetes dataset is obtained from the UCI ML Repository. The data set represents 10 years (1999-2008) of clinical care at 130 US hospitals and integrated delivery networks. The data has around 100k+ observations and 50 attributes. The data contains categorical attributes such as race, admission type, insulin, medical specialty of admitting, etc. and binary variables such as gender, change, diabetes Medicine and numeric variables like age, time in hospital, number of lab test performed, number of medication, number of outpatient, inpatient, and emergency visits in the year before the hospitalization. The target variable has 3 different classes which are “No readmission”, “A readmission in less than 30 days” & “A readmission in more than 30 days”.