54. Missing Values and Class Prediction Based on Mutual Information and Supervised Similarity

In recent times, the uses of data mining techniques have increased tremendously due to the increase in a large amount of data. Data mining techniques have been used for many research purposes. But mostly, they all face a single unique problem and that is the missing values of data. During research, large datasets are taken as processed for experimentation of algorithms, and if there is a missing value, these instances are either ignored or any default values are replaced during pre-processing of data. But this way is not correct. In this chapter, a novel prediction technique is proposed that can be used to predict the missing values of a given dataset or a dataset sample by calculating the mutual information, supervised similarity, and cosine similarity. The proposed approach calculated the missing values accurately, and this is experimented using a sample cancer dataset with missing gene values. The proposed prediction technique can also be used to predict class values of new instances of dataset. The experimentation shows that the predicted missing values and class labels coincide with the existing gene subsets and are said to be reliable and accurate.