Prediction of Saffron Flower and Stigma Yield Based on the Physical and Chemical Properties of Water and Soil Using Linear Multivariate Regression Models and M5 Decision Tree

The exciting growth of various sciences and technologies and the complexity of decision-making in recent decades, have marked new ways for humanity to use information systems and artificial intelligence, accurately and quickly and provide a solution for its time-consuming scientific and technical predictions and calculations. This study was conducted to evaluate the predictive outcome of two stepwise regression models and the M5 decision tree model under the influence of different water and soil properties on saffron flower and stigma yield in 2019 in Sabzevar saffron fields (located at longitude "57.43" Latitude "36.12") and was performed in the laboratory of the Faculty of Plant Production of Gorgan University of Agriculture Sciences and Natural Resources. In April, after the end of the saffron growing season, soil samples from a depth of zero to 30 cm and 12 irrigation water samples from 69 saffron farms were prepared and transferred to the laboratory to analysis and measuring the physical and chemical properties of soil samples. Around 13 parameters including pH, acidity, and percentage of soil components, soil elements, etc. and acidity, bicarbonate, etc. were measured in soil and water samples, respectively. Flowers were collected at the time of flower emergence from the specified areas of the fields and the desired measurements were made. The results showed that, as the soil and water analysis is relatively expensive, the M5 decision tree model has more accuracy due to the speed and lower cost than the regression model. So that, in the result of predicting the stepwise regression model, in the most ideal case and entering all the measured parameters, dry stigma weight and flower weight with correlations of 70 and 74%, respectively, and the error value is 0.23 RMSE and RMSE 16.38 were predicted. While the M5 decision tree model with lower parameters had a high capability to predict flower weight and dry stigma weight. It estimated the weight of dry stigma and flower weight with 90% correlation and error value equal to RMSE = 0.12 and RMSE = 9.4 at the end of modeling for the study area. Therefore, the M5 decision tree method is recommended in evaluating and predicting various factors on saffron yield.