Predicting Emergency Repairs using Classification Method

Abstract

This paper discusses how each explanatory variable affects the possibility of having an emergency repair to people’s home with the help of machine learning. Here, the outcome variable is binary. The aim of this is to determine whether increasing the frequency of routine repairs would decrease the frequency of emergency repairs, and the predicted probability of having an emergency repair based on the variable statuses for each property. Data exploratory is first carried out to understand and simplify the dataset obtained from a Housing Association. Statistical models such as logistic regression, decision tree, random forest, linear discriminant analysis and k-nearest neighbours are then used to fit the model to the dataset. We also investigate ways to approach the missing values. The best fitted model is then determined by comparing the highest accuracy of the predicted probabilities between these models.