Machine learning methods application to the perennial grasses yield forecast model building under irrigation

Abstract

Purpose: to test the hypothesis on possibility of perennial grasses yield forecast model building using machine learning methods. For agricultural production planning, the yield forecasting is one of the important tasks, and when using information technologies, building forecast models requires significantly less labor while maintaining or even increasing the accuracy and adequacy of the models. Materials and methods. Such methods of scientific knowledge as analysis and synthesis of various levels, methods of data preparation for machine learning, machine learning methods implementing a linear regression model and regression using decision trees were used. The popular Python language and publicly available libraries for data analysis and preparation and building machine learning models were used as the main technology stack for implementing machine learning methods. Results. The data was prepared for training the model using machine learning methods, and the data was modified without loss in such a way that it represents exclusively numerical values. Linear regression and regression models using decision trees were trained. A metric of model quality confirmed by high values of determination coefficients (0.68 and 0.98, respectively) was calculated, using standard tools from the libraries used. Conclusions: numerical values of input of such characteristics as food regime (0.176935–0.315269), as well as the ratio of the number of irrigations and irrigation norms (0.056576–0.108139), into the model, confirm their significance for the conditions of research areas and database collection.

Keywords

• machine learning
• yield forecast
• predictive model
• perennial grasses