An Ensemble Based Clustering and Classification Framework for Prediction of Agricultural Crop Yield

Abstract

One of the most important sectors of India's economy, affected by a variety of physical and climatic factors including soil quality, temperature, rainfall, and irrigation methods is agriculture. If farmers and other players in the agricultural supply chain are to make decisions based on accurate knowledge, they must be able to exactly project crop yields. On the other hand, the methods now in use find it difficult to manage the variability of agricultural fields, which leads to less than perfect accuracy and generalizability. By means of the integration of physical and climatic elements, this work presents a unique ensemble-based clustering and classification framework, so improving the accuracy of crop yield prediction. The proposed framework consists in three basic components: preprocessing agricultural datasets in order to handle missing and noisy data; ensemble clustering using Chaotic Cuckoo Search Optimization (CCSO) and Simulated Annealing (SA); and classification of clustered datasets using Support Vector Machine (SVM). This approach solves farmers' problems by pointing out significant physical and climatic signals that guide their selection of crops suitable for particular regions. Using paddy crop yield datasets, the framework was evaluated in line with Java environment implementation. For validation's purposes, performance criteria including accuracy, precision, recall, F-measure, and percentage error were used. The results show that the proposed CCSO-SA clustering ensemble with SVM classification performs rather better than the methods currently applied. Its 95.3% high accuracy, 94.8% precision, 94.1% recall, and 3.9% reduced percentage error point to one other. This progress indicates that including dependable classification algorithms and sophisticated optimization methods helps one to reach the expected results.