A novel early stage drip irrigation system cost estimation model based on management and environmental variables

Area of Science:

• Agricultural Engineering
• Environmental Science
• Data Science

Background:

• Pressurized irrigation systems, particularly drip irrigation, are crucial for water management.
• Accurate early-stage cost estimation for these systems is complex and underexplored.
• Understanding cost drivers is essential for project planning and investment.

Purpose of the Study:

• To develop and validate machine learning models for early-stage cost estimation of drip irrigation systems.
• To identify key environmental and management features influencing system costs.
• To compare the performance of various feature selection and machine learning algorithms.

Main Methods:

• A database of 515 drip irrigation projects was compiled, including 39 environmental and management features.
• Feature selection algorithms (e.g., LCA, FOA, Wrapper) were employed to identify significant cost-influencing factors.
• Machine learning models, including Support Vector Machine (SVM) and Artificial Neural Network (ANN), were trained and tested.
• Model performance was evaluated using metrics such as R-squared (R²) and Root Mean Squared Error (RMSE).

Main Results:

• The LCA and FOA feature selection algorithms demonstrated excellent estimation performance (R² ≈ 0.94, RMSE ≈ 0.002).
• For readily available features, these algorithms achieved R² of 0.95 and RMSE of 0.0006.
• The SVM model (RBF Kernel) proved most effective for overall cost estimation (R² ≈ 0.89-0.92).
• The ANN model (MLP) performed best for readily available features (R² ≈ 0.88-0.91).

Conclusions:

• Machine learning models, particularly SVM and ANN, can accurately predict early-stage drip irrigation costs.
• Environmental and management features are significant predictors of system costs.
• The developed models provide a valuable tool for accurate cost estimation in irrigation projects.