Algoritma Decision Tree of Machine Learning Dalam Klasifikasi Tanaman Kopi Agroforestri dan Kopi Monokultur dengan Citra Satelit SPOT-7 di Kabupaten Tanggamus

Abstract

Penelitian ini menjelaskan pengembangan algoritma machine learning untuk membangun model algoritma dalam mendeteksi dan mengidentifikasi tanaman kopi agroforestri dan kopi monokultur menggunakan pendekatan machine larning. Data utama yang digunakan dalam penelitian ini adalah citra satelit SPOT-7 untuk mengembangkan indeks vegetasi (NDVI, VDVI, VARI, NRGI) dan data bio-sosio-geofisik (proximity jalan, proximity sungai, proximity permukiman, elevasi, kelerengan) serta tutupan lahan. Penelitian ini menemukan algoritma terbaik untuk mendeteksi tanaman kopi agroforestri dan kopi monokultur menggunakan algoritma decision tree dengan parameter information gain dan peubah NDVI, VDVI, NRGI, elevasi, proximity jalan, proximity sungai, tutupan lahan. Algoritma ini mendapatkan maximal depth 31, tanpa pruning dan pre-prunning, minimal leaf size 51, minimal size for split 41, pre-prunning alternative 0. Algoritma ini menghasilkan overall accuracy 94,7% dan kappa accuracy 93,9%. Kopi agroforestri mempunyai user accuracy 91,6% dan producer accuracy 87,8%, kopi monokultur mempunyai user accuracy 92,7% dan producer accuracy 95,3%.

This study describes the development of machine learning algorithms to build algorithm models in detecting and identifying agroforestry and monoculture coffee plants using a machine learning approach. The main data used in this study were SPOT-7 satellite imagery to develop vegetation indices (NDVI, VDVI, VARI, NRGI) and bio-socio-geophysical data (road proximity, river proximity, settlement proximity, elevation, slope) and land cover. This study found the best algorithm for detecting agroforestry and monoculture coffee plants using a decision tree algorithm with information gain parameter with variables of NDVI, VDVI, NRGI, elevation, road proximity, river proximity, land cover. This algorithm determined maximum depth of 31, without pruning and pre-prunning, minimum leaf size of 51, minimum size for split of 41, and pre-prunning alternative of 0. The algorithm provided overall accuracy of 94,7% and kappa accuracy of 93,9%. The producer’s and user’s accuracy of agroforestry coffee were 87,8% and 91,6%, while the producer's and user's accuracy of monoculture coffee were 95,3% and 92,7%.