Predictive Analysis Using Modified Inception Net and Support Vector Machine for Forecasting Copra Yield in Agriculture

One of the major economic sectors in various states is copra agriculture, with a significant portion of its global production coming from countries such as India, Indonesia, and Sri Lanka. However, the severity of copra issues can vary depending on factors such as moisture levels, fungal presence, and the quality of the copra, leading to imbalanced data and potential overfitting in copra image analysis. Threshold-based segmentation is involved to identify different segments of copra based on factors like moisture content. Higher moisture content may indicate areas prone to fungal growth or poor drying conditions, while segments with uniform colour and texture may signify high-quality copra. The proposed approach involves using the Modified Inception Net (MIN) for feature extraction and Support Vector Machine (SVM) for hyperparameter tuning. SVM handles the kernel function in nonlinear and high-dimensional spaces, enhancing its capabilities and mitigating overfitting. However, the MIN-SVM combination may require significant computational resources, particularly during training and parameter tuning. The proposed MIN-SVM method achieves promising performance metrics, including a Mean Absolute Error (MAE) of 0.071, Mean Square Error (MSE) of 0.0125, Matthews Correlation Coefficient (MCC) of 0.620 and
$R^{2}$
value of 0.9410. This performance is compared to existing methods such as Back Propagation Algorithm (BPA), K-means and SVM.