Comparative Analysis of Supervised Classification Algorithms for Urban Sprawl Detection

Abstract

As Human population is increasing, a number of small towns are turning into big cities. But human race has developed itself in technological terms smartly, which is helping human kind to act efficiently and consume the resources appropriately. Collection of urban sprawl statistics of an area has become efficient by using remote sensing. In comparison to the traditional methods, the new method of using remote sensing for the detection and classification of urban sprawl has substantially enhanced. Using this data management has become capable of taking suitable measures for its residents. This work compares six supervised classifiers, i.e Maximum Likelihood, Minimum Distance, Support Vector Machines, Mahalanobis, Parallelepiped and Feed Forward Neural Network for urban classification. The data used is of SPOT 5 and criteria for comparison of classifiers is based on accuracy. Due to the absence of blue band in SPOT imagery the collected data samples for training tend to be complex and overlapping. Training data collected and divided into 10 different samples, show 82.74% accuracy for Artificial Neural Network. In comparison with Artificial Neural Networks, the lowest recorded results are of parallelepiped Classifier.