Combination of relief feature selection and fuzzy K-nearest neighbor for plant species identification

Abstract

Plant species identification is a digitally challenging object for a better classification such as in taxonomy resources problem. Feature selection as a preprocessing technique in data mining help to identify the prominent attributes of herbal leave with higher dimensioned data set. For this purpose, Relief Feature Selection algorithm was utilized for the improvement of Fuzzy K-Nearest Neighbor (Fuzzy K-NN) classification on shape, texture, and margins on the leaves. Best result was obtained on 73.48% of accuracy rate for 363 observation data. The trend of accuracy rate was directly imposed by the number of features. However, most of this combination was better than conventional K-NN alone.