Predicting heart disease using ant colony optimization / Siti Aisyah Ismail

The death rate due to heart disease has caused alarming concern among health experts in Malaysia as it increases year-on-year. They have to make more effort to detect heart disease, but it is not an easy task. Thus, this study used the Ant Colony Optimization algorithm with data mining called Ant-Miner to predict heart disease because it is said that Ant-Miner’s rule list is simpler than other rule induction algorithms. The aim of this study is to develop a classification model for predicting heart disease. The data set is discretised by converting the numeric attributes to the nominal attributes by using WEKA as a tool. After that, the dataset was run in the Gui Ant-Miner to find the rules and percentage of accuracy in predicting heart disease. The results of Ant-Miner’s accuracy are later compared to J48 for better classification. The dataset was run using a different number of ants from 100 to 400 to observe changes in accuracy, number of rules and number of conditions. In addition, rules and condition number were also observed when the value of the minimum case per rule was changed. The cross-validation number was set to k=10 times throughout the test due to low bias and variance, while other parameters are set with fixed value, such as maximum uncovered cases equal to 10, convergence rules equal to 10, and iteration numbers equal to 100. In conclusion, it was found that the accuracy of Ant-Miner was 78.85% while the accuracy of J48 was 73.93%, indicating that Ant-Miner had better accuracy compared to J48.