Machine Learning for Algorithm Selection

Blades, A. (2017). Machine Learning for Algorithm Selection (BEng (Hons) Software Engineering Dissertation). Edinburgh Napier University (Hart, E., Sim, K.).



This paper covers the research and implementation of two algorithm selection models. The first is built to predict a fitness score which is then used to predict the best possible algorithm. The second is built to predict the algorithm name. These predictions will be made by training machine learning models to recognise the features of a bin packing problem.

The second goal of this paper is to identify which features of a bin packing problem hold more relevance when training a machine learning model. From the fourteen features selected no feature was found to have any more relevance over another.

The algorithm selectors both worked well on the problem set used, producing final accuracies of above 85%. Analysis of the results showed that the selector that predicted a straight algorithm name produced more consistent results, but did not achieve the highest overall accuracy which has been attributed to the imbalance in the dataset.
[Read More]


Adrian Blades
Part-Time Demonstrator
+44 131 455

Areas of Expertise

Software Systems
The institute's expertise in software engineering encompasses the entire development lifecycle, and we focus on developing the approaches and tools to improve the engineering process of both emerging and widely used software systems.

Associated Projects