Research Title:  "Towards Robust Crop Yield Prediction using Stochastic Differential Equation Model using Parameter Estimation by Neural Network"

Abstract:

The prediction of wheat yield (Triticum aestivum) is a Statistically and Machine learning challenging problem due to high-dimensional data structures genotype-by-environment (G×E) and other unknown factors to predict the flowering time (heading date). We propose a stochastic differential equation model to predict the wheat flowering time based on parameter estimation. We use Multi-layer perceptron (MLP) estimation. Each genotype and Environment data feature is represented by input data multiplied by a stochastic parameter. The SDE parametric data is represented by Neural network weights. GxE parameters are calculated by identifying the NN weights however, the genetic parameters are hot encoding values represented as masked G×E corresponding NN weights. The masked GxE is implemented as a dropout technique of NN weights. The training and testing data are generated from the SDE model using different Brownian motions. The proposed stochastic model gives a promising, accurate approximation of the flowering time of a dataset during 4 years with 26 genetic markers and 44 44-environmental-feature features dataset provided by the CSIRO Agriculture and Food Dep.