A Deep Learning Approach for Dealing with Tabular Data in Crop Yield Prediction

This paper outlines the development of a deep learning approach for dealing with tabular data in crop yield prediction. The deep learning approach is developed based on the depthwise separable convolutions used in MobileNet. The results demonstrate the performance of the proposed approach outperforms the existing approaches like Multi-Layer Perceptron (MLP), Tabular Neural Network (TabNet), Residual Networks (ResNet) in terms of Validation Root Mean Squared Error (RMSE) of 930.90, Validation Correlation Coefficient of 94.39%, and R2 of 89.07. The proposed deep learning approach with its group-wise processing and iterative refinement makes it unique by allowing the model to learn inter-group dependencies, dynamically adjusts feature importance, learn feature importance at each refinement stage, and captures complex interactions.