Automatic Machine Learning for Insurance: H2O Experiment

Abstract

This thesis provides an introduction of machine learning (ML), shows the implication that ML has on the insurance sector and takes a special consideration to the H2O ensemble modelling approach for the insurance claim fraud detection binary classification. The aim of this thesis is to study the H2O Automatic ML potential and compare the results generated with traditional algorithms such as lineal perceptron, Logistic Regression, multilayer perceptron, support vector machine and decision tree. Using H2O web interface or R programming, not only the most efficient ML algorithms are obtained with no effort but also provide better modelling metrics than traditional methods.