Missing Data Estimation Using Ant Colony Optimization Algorithm

The advent of more advanced and sophisticated techniques such as deep autoencoders, stacked autoencoders, convolutional neural networks and deep convolutional neural networks amongst others, which fall under the category of deep learning, has brought about the need to investigate their effectiveness and possible use in the missing data domain. Deep learning (DL) is a branch of machine learning in the artificial intelligence arena that comprises networks which can learn and extract information from data using an unsupervised learning algorithm especially if the data is unlabelled or unstructured. Ant colony optimization (ACO) is a population-based metaheuristic algorithm capable of finding approximate solutions to complex optimization problems. In this chapter, we present a new framework for missing data imputation in high-dimensional datasets. A deep autoencoder is used in conjunction with the ACO algorithm (DL-ACO). The performance of the proposed technique is experimentally tested and compared against other existing methods of a similar nature using an off-line handwritten digits image recognition dataset.