This thesis is concerned with developing a predictive risk model to identify patients that are at high risk of readmission to hospital. Such a model should have a desirable level of predictive accuracy but also, should be financially beneficial to the DHB. Logistic regression and Naive Bayes probabilistic classification methods were both considered to build the predictive model. Performance measures such as the positive predictive value (PPV) and cost savings analysis were used to find the optimal days between initial admission and readmission and the optimal threshold for prediction of high risk patients. This study is concerned with Waikato District Health Board (DHB) domiciled patients discharged between 1 July 2009 and 31 October 2013. The dataset includes information about the patients initial admission and the response variable is whether a readmission occurred or not. Using logistic regression, this study found the model that fits the data best includes 21 variables that contain information about the patients initial admission. The two classification methods used produce a risk probability between 0 and 1 for each patient in the study. The logistic regression model performance was better than Naive Bayes as shown by the PPV (the proportion of patients correctly identified as at risk over the total at risk). The 56 day readmission data PPV at a risk threshold of 0.5 for the logistic regression was 48.4% and 30.8% for Naive Bayes. Analysis of the PPV identifies the risk threshold level of 0.5 and readmission period of 56 days as optimal predictive criteria in this study. Cost savings analysis also supports the 56 day model with an intervention cost of $500. The 0.5 cut off point in the 56 day model identifies a reasonable number of patients at risk for intervention, approximately 3,000, which equates to about 2 patients at risk per day over the period of this analysis. This analysis found the optimal model for predicting patients at risk of readmission is a logistic regression model using 56 day readmission data and a risk threshold of 0.5. A key recommendation of this study is that the DHB needs to introduce a method that correctly flags patient admissions. The model should be used on a trial basis at the DHB to see how accurate it performs.