Application of Particle Swarm Optimization to Improve the Performance of the K-Nearest Neighbor in Stunting Classification in South Sumatra, Indonesia

This research aims to obtain the best accuracy in classifying stunting children's data using K-Nearst Neighbor (KNN) by combining Particle Swarm Optimization (PSO). The K-NN algorithm is an algorithm which is an unsupervised algorithm, and is proven to be good in data mining while Particle Swarm Optimization (PSO) show. Better optimization performance compared to other methods. The methodology in this research is data collection, data pre-processing, classification of stunted children, data sharing, searching for the optimal k value to the classification process and performance testing or Particle Swarm Optimization. This dataset has an abnormal data structure where certain attribute values have quite wide ranges.The results of the K-NN classification, the average accuracy of each fold, shows that the highest accuracy was obtained at a value of k = 10, namely 86.08% and the lowest was in the last experiment with a value of k = 7500 of 72.67%. It can be concluded that the higher the k value, the resulting accuracy will increase. Meanwhile, the results of K-NN classification with PSO can be concluded that the higher the w value, the greater the possibility of getting better fitness. This result is also in accordance with research where the best w value is above 0.5 and less than 1. This is because if the w value is more than 1 it can cause the particles in the PSO to become unstable because the resulting speed is not controlled. It is proven from the test results that the range This value produces better average accuracy and starts to decrease again when entering the value w = 1. Then the test results also show that a small value of w can result in the role of particle speed becoming insignificant and can increase the possibility of early convergence. It can be seen from the results of testing the number of PSO popsizes that the highest average accuracy was 93.2% at a value of w = 0.9. From the description above, KNN shows an accuracy of 86.08%, while KNN with PSO increases to 93.9%, so this shows that KNN with PSO is more accurate in classifying stunted children.