The Future of Work: Machine Learning and Employment

29 November 2015

View Journal Article / Working Paper

This project investigates the susceptibility of jobs to computerisation and particularly which features of a job determine the probability of computerisation. This is achieved by using Gaussian Process Classification. A set of labelled occupations is used to train and test the model and the effect of using different feature sets is explored. Feature selection in the form of greedy selection is used to find the feature set that achieves the best classification and thereby find the features that are most significant when determining if a job can be computerised. It is concluded that the most important feature is Originality and the best feature set for classifying the data in this project consists of Originality and Service Orientation. Furthermore, experiments are performed using linear embedding methods for feature learning. However, these experiments fail to prove that better classification can be achieved using this method.