ABSTRACT
In this chapter, we investigate the issue of classifying data whose values are not certain. Similar to other data mining solutions, most classification methods (e.g., decision trees [27, 28] and naive Bayes classifiers [17]) assume exact data values. If these algorithms are applied on uncertain data (e.g., temperature values), they simply ignore them (e.g., by only using the thermometer reading). Unfortunately, this can severely affect the accuracy of mining results [3,29]. On the other hand, the use of uncertainty information (e.g., the derivation of the thermometer reading) may provide new insight about the mining results (e.g., the probability that a class label is correct, or the chance that an association rule is valid) [3]. It thus makes sense to consider uncertainty information during the data mining process.
