Penerapan Algoritma MKNN pada Data Historis Gempa Bumi yang Berpotensi Tsunami

Abstract

In the implementation of modified K-Nearest Neighbor method for the determination of tsunami potential by comparing the calculation of euclidean distance and Manhattan used 3 earthquake criteria namely strength, depth and epicenter with 2 classification classes, potentially tsunami and not potential tsunami. The MKNN algorithm works by retrieving a number of nearby K data (its neighbors) as a reference to determine the class of new data. This algorithm classifies data based on similarity or similarity or proximity to other data. The result is that algortima can classify the status of an earthquake whether it has the potential for a tsunami or not by paying attention to the balance of the composition of the training data used. With the highest accuracy value of 90% for K=1. The results of the euclidean and manhattan comparisons were obtained from several test scenarios, namely changes in the composition of datasets, changes in the value of K and changes in training data. Obtained average system performance of 80% for euclidean distances and 82% for manhattan distance test results showed that the composition of the dataset greatly influenced the performance of the system obtained. So Manhattan has a higher level of accuracy than the euclidean distance with an average difference of 2%.