As explained earlier, the logical approach to handle missing data depends upon how data points become missing. As indicated in earlier chapters, Little and Rubin (1987) as well as Rubin (1978) have demonstrated that there are three types of missing data mechanisms and these are: Missing Completely at Random, Missing at Random and Missing Not At Random. Depending on the mechanism of missing data, currently various methods are used to treat missing data. More information with detailed discussions on the various missing data estimation methods used to handle missing data can be found in Al-lison (2000); Rubin (1978); Little and Rubin (1987); Mohamed and Marwala (2005); Leke, Marwala, and Tettey (2006); and Nelwamondo (2008). In this chapter, as in earlier chapters, the method for estimating missing data is able to estimate missing data irrespective of the missing data mechanism as long as the rules that describe inter-relationships in the data are known.

The missing data estimation algorithm considered in this chapter involves a neural network which is trained to recall itself and is, therefore, called an autoassociative neural network. Successful deployment of autoassociative neural network include that by Pomi and Olivera (2006) who developed a context-sensitive autoassociative memories and applied this formedical diagnosis, in object recognition (Caldara & Abdi, 2006; Yokoi et al., 2004), in nuclear engineering (Marseguerra, Zio, & Marcucci, 2006), in mechanical engineering (Marwala & Chakraverty, 2006), in fault detection of gearboxes (Del Rincon et al., 2005) and in spotting consonants in speech (Gangashetty,...