Rice phenology classification based on random forest algorithm for data imbalance using Google Earth engine

Abstract

The agricultural sector is one of the important sectors in the world and has a very significant contribution to the achievement of the Sustainable Development Goals (SDGs) program. In the SDGs, attention to food security is focused on the second key indicator namely zero hunger (SDG 2). Availability of accurate rice production data is required to measure the level of food security. This can be done by monitoring the growth phase of a food plant which is called rice phenology classification. We used Random Forest (RF) algorithm on the Google Earth Engine (GEE) platform in Lamongan Regency, East Java in 2019 to classify rice phenology from Landsat-8 satellite imagery, which has the characteristics of imbalance case. To deal with the imbalanced issue, an oversampling technique was used for sampling minority classes. Reference data for the classification model training were collected from the Area Sampling Framework survey published by Statistics Indonesia in 2019. The results showed that the overall accuracy (OA) using RF algorithm by modifying the dataset using oversampling was 81.46% and the kappa statistic (κ) was 0.76, outperforming the RF technique without oversampling.