Abstract
The research study estimated the impact of season, maturity and temperature inside
different material covers in predicting defect on banana peel from hand weight (HW), pulp
per peel ratio (PPR), sunburn (SB) as well as thrips damages (TD) score of Cavendish banana
in Chiang Rai. The banana bunch covers were conducted in summer (May to June 2017)
and in rainy (August to October 2017). A one hidden layer feed forward backpropagation
artificial neural network (ANN) was developed by varying the hidden node in a hidden layer
for 2-20 nodes. Four separately ANN models were performed for Cavendish banana to
predict qualities by using R and RStudio programs. Data input variables were rate of heat
energy transmitted (Qx), hand location, and temperature profiles. The results showed that
the 4-18-1, 6-16-1, 5-8-1, and 5-12-1 architectures were the most suitable model for HW,
PPR, SB and TD score, respectively. The model performance presented the relatively high
R2 of 0.76, 0.96, 0.86, and 0.88 for HW, PPR, SB and TD score, respectively. Moreover, the
selected model provided root mean square error (RMSE) of 1,157.62g, 0.252, 8.407, and
1.310 for HW, PPR, SB and TD score, respectively. The computational model ANN was
used for the prediction of hand weight, pulp per peel ratio (maturity), sunburn, and thrips
damages for Cavendish banana production in Thailand.
