Comparison of Predictive Performance of Data Mining Algorithms in Predicting Tomato Yield with the A Case Study in Igdir

Abstract

Among the vegetable species in the world, the plant with the most cultivation area is tomato. Increasing tomato yield is important in terms of contributing more to the world economy, producer’s income and human health. With the advancement in software technologies, the importance of data mining algorithms is increasing due to the fact that these algorithms can produce more sophisticated solutions for regression and classification problems. Determining the factors affecting tomato yield and comparing different data mining algorithms on prediction of tomato yield are the purpose of this study. For this purpose, survey study was conducted with the 105 farmers, selected by Simple Random Sampling Method in Igdir province in 2016. Different data mining algorithms including Classification and Regression Tree, Exhaustive CHAID, Chi-Square Automatic Interaction Detector, Artificial Neural Network Algorithm, Multivariate Adaptive Regression Splines and General Linear Model were developed and compared their predictive performance. MARS decision tree has been able to build a model with greatest predictive accuracy, and the others are respectively ANN, GLM, CART, CHAID and Exhaustive CHAID. In the MARS model, number of irrigation , amount of chemical fertilizer , age of farmer , number of seedlings , education level , soil analysis status , sowing region were found statistically significant (P˂0.05). Preferring the MARS model could give an opportunity to detect factors affecting tomato yield and their interactions with higher accuracy. Moreover, results can be easily interpreted and the rules are understandable.