MODEL FUZZY K-NEAREST NEIGHBOR WITH LOCAL MEAN FOR PATTERN RECOGNITION

K-Nearest Neighbor is one of the top 10 algorithm in data mining (Wu, 2009). Based on its development, K-Nearest Neighbor is combined with Fuzzy's approach. Fuzzy K-Nearest Neighbor located as membership degrees - except euclidean distance - as a feature of the data attachment to the target class so that Fuzzy KNN is known to improve the classification results. Except for adding Fuzzy, K-Nearest Neighbor is also modified at the class determination stage with Local Mean rules. At Local Mean KNN, the value of the data vector's test were calculated in each target class so that the euclidian distance was not calculated between the data but it was also between the target classes. In this study, we divide the local mean vector of LMKNN by the degree of membership for each class produced by Fuzzy K-Nearest Neighbor to obtain a smaller value vector. This will affect the more obvious range of values of the trend of a data to a class than other class. The test was performed using Iris dataset with k taken as many as 3 nearest neighbors in each target class. Accuracy results obtained with data testing in each class are 93.3%, 86.6% and 100%, so the overall average is 93.3%.