IDENTIFIKASI SPESIES MANGROVE MENGGUNAKAN ALGORITME RANDOM FOREST

Abstract

thm, that are preprocessing (SMOTE) and min-max normalization to balance the data distribution. The result shows that the projection of the normalization range (interval 0-1) has no effect in reducing the data pattern dimensionally. After preprocessing and normalization, five attributes (species, wood density, diameter at beast height, total of above ground biomass, and below-ground root) were classified and analyzed with species as the target attribute. The construction of model parameters is based on the total number of SMOTE results by specifying 100 and 500 as the number of single tree and 1000 as the number of nodes and the default predictor variable. The final result shows that the Random Forest algorithm obtained an optimal evaluation value with an average of 99.97% using the number of single tree and the predetermined cut-offs. The maximum accuracy of 100% is obtained from the number of single tree and cut-off with the following sizes: (1) 500 and 80:20; (2) 500 and 90:10; and (3) 100 and 80:20. These results indicate that it is very effective to apply the Random Forest algorithm as a machine learning classifier method to classify mangrove species. Keywords: