Document Type : Research Article
Authors
1 Ph.D. Candidate, Department of Agricultural Economics, University of Tabriz
2 Dep. Agricultural Economics, University of Tabriz
3 University of Tabriz
4 Marie Sklodowska-Curie Postdoctoral Research Fellow at Fondazione Eni Enrico Mattei, Milan, Italy
Abstract
Introduction
Optimizing agricultural resources is crucial to meeting the food demands of a growing population. This involves increasing cultivated land or maximizing production efficiency, especially in the face of challenges like climate change and water scarcity. Evaluating the efficiency of agricultural producers helps identify gaps between top performers and others, which can inform policies aimed at enhancing productivity. Ahar County is a key wheat production area, contributing about 10% of East Azerbaijan Province's wheat cultivation, with 28,000 out of 52,000 hectares dedicated to this crop. While this indicates the county's significance, the varying yields necessitate further investigation into technical efficiency (TE).
Conducting economic research on wheat production in Ahar could positively impact regional yield. This study aims to provide a framework for analyzing factors affecting efficiency, aiding farmers in making informed resource management decisions. Furthermore, data mining techniques can analyze economic, social, and environmental data to uncover complex relationships influencing efficiency, thereby enhancing production optimization and guiding policymakers in improving agricultural productivity.
Material and methods
In this study, data mining techniques are employed to classify farmers based on TE and to analyze the factors influencing it. First, TE is calculated using wheat output (kg) and inputs such as cultivated area (hectares), seeds (kg), labor (person-days), and tractor usage (hours) through Data Envelopment Analysis (DEA).
Farmers are classified into two groups: High-efficiency farmers (above regional average efficiency) and low-efficiency farmers (below average efficiency). Next, the most suitable machine learning algorithm is identified to predict farmers' TE and categorize them accordingly. The aim is to find an algorithm that closely predicts the actual classifications based on efficiency calculations. Higher algorithm accuracy improves the reliability of the analyses. Machine learning, particularly supervised learning algorithms, is used for this analysis. Classification algorithms aim to create a model for predicting outcomes by sorting database records into predefined categories based on specific criteria. Common tools include logistic regression, support vector machines, k-nearest neighbors, and random forests.
After estimating machine learning models, it is essential to evaluate their performance using various metrics to ensure the models function correctly. Key metrics include the confusion matrix, recall, accuracy, precision, F1 score, Cohen’s kappa statistic, and the Receiver Operating Characteristic (ROC) curve, each highlighting different aspects of model performance. Utilizing these metrics allows for a more detailed analysis of predictive results. The data used in this study were collected through simple random sampling via in-person interviews and questionnaires from 223 wheat producers in Ahar County.
Results and discussion
The average TE was calculated to be 0.59. Farmers with an efficiency below this value were categorized into group zero, while those with an efficiency above this value were placed in group one. The results of the statistical tests indicated that continuous variables such as nitrogen fertilizer, land rental value, age, and experience; countable variables like the number of plots and household members with university education; dummy variables including residence, seed sourcing, land ownership, harvesting method, off-farm income, animal manure, herbicide, and pesticide; and weed control, as the only ordinal variable, significantly affect technical efficiency and should be included in the machine learning model.
The study utilized several algorithms, including logistic regression, support vector machine, K-nearest neighbors, and random forest. Logistic regression achieved the highest average accuracy of 88.7% with five-fold cross-validation and outperformed the others, showing an AUC of 0.98 on ROC curves and strong performance in the confusion matrix. The results of the logistic regression indicate that variables such as herbicide use, weed control, animal manure, land rental value for wheat farms, nitrogen fertilizer, number of farm plots, pesticide use, age, combined harvesting methods, experience, household members with university education, seed supply from the Agricultural Jihad Organization, and living in rural areas positively affect TE. Conversely, land ownership (personal-rented), sourcing seeds from personal resources, and having off-farm income negatively impact efficiency.
Optimal herbicide application can reduce competition with wheat, enhance crop quality, and improve nutrient and rainfall absorption, thereby increasing TE. Using animal manure improves soil quality, aiding wheat's nutritional needs. High land rental values may indicate suitable soil quality and production potential, facilitating modern management practices that positively influence efficiency. Nitrogen fertilizer can enhance plant growth and yield by increasing soil nitrogen levels. Farmers with multiple plots can mitigate risks from weather or pests and employ better planting and harvesting strategies. Additionally, pesticides help prevent damage from pests, increasing overall production. Age often reflects a farmer's experience in farming, with older farmers typically having more expertise in effective agricultural practices, which can improve TE. Experienced farmers tend to adopt better strategies to address agricultural challenges. Using combined harvesting methods can optimize the harvesting process, allowing better use of time and resources and resulting in higher yields. The presence of educated members within farming households can enhance agricultural management knowledge and skills. Sourcing seeds from reputable institutions like the Agricultural Jihad Organization, which provides quality and pest-resistant seeds, can effectively enhance crop yield. Living in rural areas is a crucial factor for timely agricultural operations.
On the negative side, land ownership (personal-rented) can reduce farmers' incentives to invest in land improvement and resource optimization, ultimately impacting their TE. Sourcing seeds from local personal resources can lead to lower quality and pest-sensitive seeds, resulting in reduced productivity. Additionally, off-farm income may divert farmers' focus from agricultural activities, diminishing attention and investment in farming practices, which can lead to decreased TE.
Conclusion
The study suggests that training programs and sourcing seeds from reliable suppliers, like the Agricultural Jihad Organization, can enhance farmers' yields and crop quality. However, land ownership (personal- rented) and sourcing seeds from personal resources may reduce efficiency. Policymakers should motivate farmers on rented land with financial incentives and improve access to quality seeds through distribution centers in rural areas. Collaborations with seed-producing companies can ensure a steady supply of high-quality seeds, further boosting yields.
Keywords
Main Subjects
)
Send comment about this article