Machine learning approach to predict susceptible growth regions of Moringa peregrina (Forssk)

• Autor(a) principal: Moradi, Ehsan
• Data de Publicação: 2021
• Outros Autores: Abdolshahnejad, Mahsa, Borji Hassangavyar, Moslem, Ghoohestani, Ghasem, da Silva, Alexandre Marco [UNESP], Khosravi, Hassan, Cerdà, Artemi
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1016/j.ecoinf.2021.101267; http://hdl.handle.net/11449/207473
• topic: Classification and regression trees; Moringa peregrina; Multivariate discriminant analysis; Random forest; SDGs; Support vector machine

Moving towards sustainable products and services in regions with fragile ecosystems needs plant species such as Moringa peregrina (Forssk) that will contribute to the restoration of the land and the development of the societies. This tree species is known as a source of income for local people via preparing medicine, food, industrial oil, livestock feed, and an effective role in water and soil conservation. In recent years, the reduction of M. peregrina has damaged ecosystem services in south-eastern Iran. According, the main objective of this study is to use new Machine Learning (ML) models include: Support Vector Machine (SVM), Multivariate Discriminant Analysis (MDA), Random Forest (RF), and Classification and Regression Trees (CART) to predict the regions susceptible to M. peregrine recovery. South Baluchistan in Iran was selected as a study area due to its location in a represent amen region where sustainable environmental production is threatened by land degradation processes. The location of 83-plant mass of M. peregrina was recorded in field visits by a global positioning system (GPS) device to recognize the relationship between them and thirteen meteorological, morphometric, and geological indicators. Within the 83 selected sites, 70% of them were used for training and 30% used for ML models calibration to predict the susceptible growth regions of M. peregrina to determine the most important indicators affecting his presence and to determine the prediction accuracy for ML models, the Jackknife test method and the area under the receiver operating characteristics curve (AUC) were used, respectively. The results showed that rainfall was the key indicator that determines the success of the plant establishment. So that, it had the most value of the percentage of relative decrease (PRD) as the following was 20.68, 30, 24.52, and 14 for the SVM, MDA, RF, and CART models, respectively. Models validation showed that the RF model with an AUC value of 0.882, is an efficient and reliable model to predict the regions susceptible to growth M. peregrina. It followed by the CART (0.849), MDA (0.832), and SVM (0.827). The final map of the RF method demonstrated that the area with a higher probability for growing M. peregrina is the wettest one. The results of this investigation are the potential map of M. peregrina growth that will contribute to the restoration of the land and will increase primary production, water, and soil protection, increase local people's income and achieve the Sustainable Development Goals (SDGs).