Efficient identification and follow-up of astronomical transients is hindered by the need for humans to manually select promising candidates from data streams that contain many false positives and prioritise those candidates for follow-up. These difficulties are faced by most ground-based time domain surveys. This dependence on humans is unsustainable for next generation all-sky surveys and significant effort is now being invested to solve the problem computationally. In this work we investigate machine learning approaches to difference image artefact rejection and contextual classification of transient discoveries. We construct training sets from data gathered during the course of Pan-STARRS1 survey operations. In all cases the feature representation is carefully selected such that it is independent of the specific Pan-STARRS1 image processing pipeline and survey strategy, with the aim of designing solutions that can be easily applied to other current and future transient searches. We explore a number of machine learning algorithms and carefully evaluate their performance on these tasks. This thesis has developed working code that is now applied on the live PS1 data stream on a nightly basis. This acts as a useful testbed for application to future surveys such as LSST.