Label ranking‐based recommendation system to rank crops for agroecological units

Abstract

Data mining and machine learning techniques help to predict suitable crops in ranking order for a given location with its soil nutrient status. We proposed a tuned bagging‐based K‐nearest neighbor ensemble label ranker and it is used to predict the ranked crops for density‐based spatial clustering of applications with noise (DBSCAN) based compressed crop‐ranked agricultural data. This ensemble integrates the commonly used Borda rank aggregation method, but there is a possibility to improve the performance of our proposed algorithm by selecting the most suitable rank aggregation based on the improvement in the label ranked dataset. The results of the study show that Copeland performed better than Borda aggregation with a 6.65% improvement in the performance of our algorithm, while it is 5.16% for Borda. The voting rule selection (VRS) is also integrated into our study with dataset‐level and instance‐level learning on crop‐ranked datasets. VRS and Copeland aggregation methods shared the first rank position with 46.14% and 45.67% improvement in the ranked datasets. The instance‐level winning percentage of crop ranked data for VRS, Copeland, and Borda is 61.44%, 56.23%, and 47.81%, respectively. This study has observed that in the case of a tuned K‐nearest neighbor label ranking algorithm, the VRS and Copeland voting rule performs better than Borda.