Comparison of Prediction Models for Arabica Coffee Berry Borer with Machine Learning in Chiang Mai and Chiang Rai

The objective of this research was to compare models for predicting Arabica coffee berry borer by machine learning, which were support vector regression technique, artificial neural network and the random forest model in Chiang Mai Province and Chiang Rai Province. The study areas were Pa Miang Royal Project Development Center, Tape Sadet Sub-district, Doi Saked District, Chiang Mai, Teen Tok Royal Project Development Center, Huay Kaew Sub-district, Mae On District, Chiang Mai, Huay Pong Royal Project Development Center, Mae Jae Dee Mai Sub-district, Wieng Pa Pao District, Chiang Rai, and Huay Nam Khun Royal Project Development Center, Ta Kau Sub district, Mae Sa Rauy District, Chiang Rai. Data, the number of coffee berry borer (amount/1,800 square meters) and monthly climate data from weather stations in each area, including temperature, maximum temperature and lowest temperature (degrees Celsius), relative humidity (percent), wind speed (meters per second) and rainfall (millimeter per day) were collected from May 2012 - February 2013 (10 months) to build the model and from March 2013 - May 2013 (3 months) for comparing models. It had been found that supports vector regression is the most effective in predicting the coffee berry borer with the lowest square root of the mean error (41.1784).