A Prediction Model Optimization Critiques through Centroid Clustering by Reducing the Sample Size, Integrating...

Area of Science:

• Agricultural Science
• Data Science
• Machine Learning

Background:

• Accurate crop productivity modeling is vital for sustainable agriculture.
• Machine learning algorithms offer advanced capabilities for data analysis and prediction.
• Traditional statistical models have limitations in capturing complex agronomic factors.

Purpose of the Study:

• To compare the performance of machine learning models (MLM) against traditional statistical models (TSM) for predicting wheat productivity.
• To optimize MLMs by integrating them with statistical approaches and generating optimized datasets.
• To evaluate the effectiveness of Random Forest Regression (RFR) in wheat yield prediction.

Main Methods:

• Generated three optimized datasets (D1, D2, D3) from an initial dataset (D1) of 26,430 crop-cut experiments.
• Integrated MLMs with traditional statistical models (TSM) including Multiple Linear Regression (MLR).
• Deployed Decision Tree Regression (DTR) and Random Forest Regression (RFR) using 75% training and 25% testing data, evaluated with R², RMSE, AIC, and E.R.

Main Results:

• MLR within MLM outperformed TSM, and both model types showed improved performance on generated datasets.
• Random Forest Regression (RFR) demonstrated superior performance across all generated datasets (D1, D2, D3, D4).
• Optimized datasets enhanced the predictive capabilities of both MLM and TSM.

Conclusions:

• Machine learning models provide a robust and effective alternative to traditional statistical modeling for predicting wheat productivity.
• The integration of MLMs with optimized datasets significantly enhances prediction accuracy.
• Random Forest Regression is a highly effective algorithm for wheat yield forecasting in agricultural contexts.