Optimum Machine Learning Algorithm with Object-Based Image Analysis for Detecting Incompliant Land Utilization in Agricultural Land Reform Areas in Thailand

Abstract

The Agricultural Land Reform Office (ALRO) was established to address the country’s developmental challenges by implementing land consolidation programs, which allocate land for both agricultural and residential use to farmers. Currently, it has been found that some land use under ALRO’s responsibility does not comply with the Agricultural Land Reform Act, including hotels, resorts, and accommodations. The primary objectives of this study are (1) to determine the optimal machine learning algorithm—among support vector machines (SVM), random forests (RF), decision trees (DT), Naïve Bayes (NB), and K nearest neighbor (KNN)—for detecting incompliant land utilization in the modeling area (Wang Nam Khiao district) and (2) to validate an optimum machine learning algorithm for detecting incompliant land utilization in the test area (Pak Chong district). The research methodology, which includes eight significant steps, is implemented by applying object-based image analysis (OBIA) combined with machine learning to classify land use from Sentinel 2A imagery in the modeling area and to validate the results in the test area. The results showed that the most suitable machine learning algorithm for detecting incompliant land utilization at ALRO 4-01 plots in the modeling area was RF, as it achieved higher overall accuracy and Kappa coefficient values than SVM, DT, NB, and KNN. The derived overall accuracy and Kappa coefficient of RF were 87.45% and 79.57%, respectively. Furthermore, the selected optimal object features and algorithm from the modeling area were effectively transferred for land use classification and the detection of incompliant land use at ALRO 4-01 plots in the test area, yielding acceptable validation results. These findings can support future monitoring and enforcement of policies in ALRO-4-01 areas.