Determination of Cassava Drought Resistance Factors by a Neural Network Method
Due to global warming, droughts are more severe and happen more frequently. Without enough water, cassava grows smaller, ultimately leading to lower market price. We consider the field data from the Department of Agronomy, Faculty of Agriculture at Kasetsart University. They study the arid resistance of cassava and analyze its morphological adaptabilityof 5 local Thai varieties as check varieties and 248varieties imported from Center of Tropical Agriculture(CIAT)germplasm.The width and length of leaves, leaf greenness, height of stems (with and without leaves) are measured. The data was collected for eight, ten, and twelve monthsafter planting. Their aim is to identify varieties with the best arid resistance.Currently, drought resistanceis compared by the K-means clustering method. K-means weights of each variable vary according to individual researcher’s judgments. As a result, the choice of the best varieties may be inconclusive. To be less subjective, we apply the Neural Network (NN) to select the best varieties and identify predicting factors for arid resistance. We found that the total height of its stem correlate with height with leaves and height without leaves, so the latter two variables are deleted.Therefore, the NN response is cassava weight, and12 independent variables are as follows: the width and length of its leaves, the height of its stem and leaf greenness of the 8th, 10th, and 12th months. We found that the Mean Square Error in the testing set is minimum when the number of nodes hidden layer is8. OurNN Method identifies the breeds with the top 5 highest weight are Mlnd22, CMN196, MEuc33, MVen134 and Mper378, respectively.The predicting factors for cassavas weightsare the width of leaves, leafgreenness and the height of its stem.
Keywords - Cassava, Neural Network, Drought, Arid Resistance