Scholarlink

<p>Groundwater plays a vital role in meeting agricultural water demands, particularly in semiarid and rapidly urbanizing regions such as Tirupati, Andhra Pradesh, India. However, increasing reliance on groundwater, coupled with declining quality due to salinization and anthropogenic activities, poses significant risks to sustainable irrigation practices. In this context, accurate assessment and classification of groundwater quality are essential for informed water management and crop planning. This study evaluated the irrigation suitability of groundwater across Tirupati Rural, Chandragiri, Renigunta, Yerpedu, and Ramachandrapuram. In total, 200 samples were systematically collected during the pre-monsoon period (January–March 2024) to minimize rainfall-induced recharge and ensure temporal consistency. To reduce diurnal variation, all samples were collected early each morning before groundwater extraction for field use. Of these, 150 samples were used for model training and testing, while 50 were reserved for validation. Physicochemical analysis, conducted in accordance with WHO and Bureau of Indian Standards, focused on 12 key parameters including pH, electrical conductivity, major ions, boron, and total dissolved solids. Although most samples met drinking water standards, only 35% were suitable for irrigation, with the remaining 65% failing due to high salinity and elevated sodium adsorption ratio. To improve classification accuracy, a hybrid random forest–genetic algorithm (RF–GA) model was developed and benchmarked against conventional machine learning algorithms: RF, support vector machine, artificial neural network, logistic regression, and Gaussian process regression. Model training employed k-fold cross-validation with training ratios ranging 60–90%, and recursive feature elimination for feature optimization. The RF–GA model outperformed all the other models, achieving a minimum cost of 1.1524% at the 41st iteration. At 80% training ratio, it achieved 97.11% accuracy, a 96.29% F-measure, and a Matthews correlation coefficient of 94.86%. The integration of the genetic algorithm enhanced global search capability and reduced premature convergence. The findings demonstrate the utility of hybrid evolutionary models in groundwater quality assessment, offering a robust decision-support tool for sustainable irrigation and regional water resource management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive Analytics for Irrigation Water Quality: An Optimized Approach by Using Genetic Algorithm

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Nerusupalli Dinesh Kumar Reddy,
  • Diksha,
  • K Praveen

摘要

Groundwater plays a vital role in meeting agricultural water demands, particularly in semiarid and rapidly urbanizing regions such as Tirupati, Andhra Pradesh, India. However, increasing reliance on groundwater, coupled with declining quality due to salinization and anthropogenic activities, poses significant risks to sustainable irrigation practices. In this context, accurate assessment and classification of groundwater quality are essential for informed water management and crop planning. This study evaluated the irrigation suitability of groundwater across Tirupati Rural, Chandragiri, Renigunta, Yerpedu, and Ramachandrapuram. In total, 200 samples were systematically collected during the pre-monsoon period (January–March 2024) to minimize rainfall-induced recharge and ensure temporal consistency. To reduce diurnal variation, all samples were collected early each morning before groundwater extraction for field use. Of these, 150 samples were used for model training and testing, while 50 were reserved for validation. Physicochemical analysis, conducted in accordance with WHO and Bureau of Indian Standards, focused on 12 key parameters including pH, electrical conductivity, major ions, boron, and total dissolved solids. Although most samples met drinking water standards, only 35% were suitable for irrigation, with the remaining 65% failing due to high salinity and elevated sodium adsorption ratio. To improve classification accuracy, a hybrid random forest–genetic algorithm (RF–GA) model was developed and benchmarked against conventional machine learning algorithms: RF, support vector machine, artificial neural network, logistic regression, and Gaussian process regression. Model training employed k-fold cross-validation with training ratios ranging 60–90%, and recursive feature elimination for feature optimization. The RF–GA model outperformed all the other models, achieving a minimum cost of 1.1524% at the 41st iteration. At 80% training ratio, it achieved 97.11% accuracy, a 96.29% F-measure, and a Matthews correlation coefficient of 94.86%. The integration of the genetic algorithm enhanced global search capability and reduced premature convergence. The findings demonstrate the utility of hybrid evolutionary models in groundwater quality assessment, offering a robust decision-support tool for sustainable irrigation and regional water resource management.