Optimizing model and feature selection using evolutionary swarm intelligence for image classification

Model selection is the way towards choosing one machine learning model among an assortment of competitor models for a training dataset. The correct model selection is vital in academic machine learning research as well as in many industrial settings. Figuring out the relevant features from a set of features and removing the irrelevant or less vital features which no longer make contributions to the target variable so that we can gain higher accuracy for our model is a common problem. Feature Selection is one of the main things which highly influences the performance of the model. The features that are used to train the models have a big effect on the performance one could gain. Irrelevant features can negatively impact model performance. Evolutionary algorithms are known for optimizing the process. In this paper we use an evolutionary algorithm called Ant Colony Optimization (ACO) to select an optimized model for given data and then use another algorithm called Particle Swarm Optimization (PSO) to select best features so that we get best of both model and features for the give data. We found that the combination of an optimal architecture selected by ACO and optimal features selected by PSO gives higher accuracy.