&npsp &npsp &npsp Last step is the core of the noise reduction: a self-adaptive decision tree classifier coupled with feature sets that were experimentally proven to yield the most accurate results in the illness classification problem. A message is first stripped of all special characters and URL’s, leaving the text to be classified by the algorithm. Then, the text is tokenized and individual weights of tokens are calculated on the based set of rules for feature sets, e.g. indication of ownership “I got a flu”. Healthify constantly strives to perfect its analytical system by analyzing large amounts of twitter data. We are improving our algorithms by utilizing human intelligence: Amazon Mechanical Turk and the ability for Users to “delete” irrelevant tweets in the HealthiFind. In total, Healthify has analyzed nearly 12 million tweets and counting!