An algorithm for creating semi-synthetic datasets for regional classification of cotton varieties

In this study, an algorithm for synthetic expansion of training datasets was developed and applied to the regional classification of cotton varieties recommended for cultivation in the Republic of Uzbekistan. The algorithm is based on heuristic logic, and the class objects were restructured based on similarity criteria. Initially, a space of textual, nominal, and quantitative features was formed using real data from the state register. Subsequently, the features were fully converted to nominal form, and the degrees of similarity between objects were determined through scaling. A proximity function and decision rules were developed, and the contribution of class objects to their respective classes was evaluated. Artificial objects were generated based on heuristic criteria, increasing the number of classes and their elements. This approach significantly improved the stability and accuracy of the classification model. Experimental results showed that the proposed algorithm achieved a precision of 95.8 %, which is substantially higher than that of the decision tree (87 %) and KNN (84.4 %) algorithms, demonstrating the effectiveness of the proposed method. The research results have created the opportunity to accurately classify cotton varieties by region.