Weather-based forecasting model for the presence of Metisa plana in oil palm plantation using feature selection in artificial neural network

Abstract

Bagworm is the most important insect defoliator of oil palm. The bagworm larvae scrape off the leaflets’ epidermis while the older larvae chew the leaflets and leaving multiple holes and causes the palm to lose its photosynthetic capability. A bagworm census should be carried out quickly to determine the extent of damage. However, the conventional practices are heavily dependent on in-situ data collection, which is destructive, less efficient, laborious, and costly. Recently, many studies have incorporated machine learning analysis such as artificial neural network (ANN) in agricultural fields especially in the development of pest prediction model. Therefore, this study was conducted to develop a weather-based bagworm prediction model using ANN-Feature Selection method. Bagworm censuses were done by identifying Metisa plana’s larval stage 1 (L1) to 7 (L7) from 13 random palms by cutting off frond number 17 biweekly and weather data was recorded by installing weather station in an oil palm plantation belongs to TH Plantation Berhad in Muadzam Shah, Pahang, Malaysia from July 2016 to June 2017. The results revealed that the significant weather parameters were frequent at time-lag 12. All the larval stage prediction models from ANN-Feature Selection were able to produce satisfactory R2 values ranging from 0.526 to 0.995. The best model was the L1 model with R2 value of 0.985 and the accuracy of more than 90 %.