Progress is being made in applying expert systems and artificial intelligence to disease diagnosis. This paper describes an algorithmic approach to improving the ability to determine whether a person has diabetes. The approach improves upon the artificial immune recognition system 2 (AIRS2) algorithm, which itself improved upon the original artificial immune recognition system (AIRS). The modified AIRS2 (MAIRS2) replaces the “k-nearest neighbors algorithm with the fuzzy k-nearest neighbors [in an attempt] to improve the ... accuracy of diabetes [diagnoses].”

The authors applied MAIRS2 to the public Pima Indian Diabetes dataset of the National Institute of Diabetes and Digestive and Kidney Diseases. MAIRS2 had a better classification accuracy (89.10 percent) than either AIRS (79.22 percent) or AIRS2 (82.69 percent), although it was not quite as good as another method, smooth support vector machine (93.2 percent). However, the approach taken with the new algorithm was validated.

The paper targets domain experts in disease diagnostics analytics who specialize in diabetes. However, by illustrating progress in the process of diagnosis for one very specific disease, the authors encourage the further use of analytics (such as those described in this paper) to improve the diagnosis of other diseases. This widens the audience of possible readers to anyone interested in the benefits of using analytics for disease diagnosis.