Scalability Analysis of ANN Training Algorithms with Feature Selection

The advent of high dimensionality problems has brought new challenges for machine learning researchers, who are now interested not only in the accuracy but also in the scalability of algorithms. In this context, machine learning can take advantage of feature selection methods to deal with large-scale databases. Feature selection is able to reduce the temporal and spatial complexity of learning, turning an impracticable algorithm into a practical one. In this work, the influence of feature selection on the scalability of four of the most well-known training algorithms for feedforward artificial neural networks (ANNs) is studied. Six different measures are considered to evaluate scalability, allowing to establish a final score to compare the algorithms. Results show that including a feature selection step, ANNs algorithms perform much better in terms of scalability.