Development of a deep learning-based expert system for early detection of corn diseases using the TF-IDF and Multi-Layer Perceptron approaches

Corn is a vital agricultural commodity, yet early disease detection remains challenging due to the linguistic variability in describing symptoms. To address this, this study proposes a deep learning-based text classification model designed to diagnose corn diseases using descriptive symptom inputs. The research contribution is the development of a robust intelligent system capable of accurately interpreting natural language symptom descriptions to overcome the limitations of rigid rule-based diagnostics. The methodology employs Term Frequency-Inverse Document Frequency (TF-IDF) for feature extraction combined with a Multi-Layer Perceptron (MLP) architecture. To ensure model robustness and generalization, data augmentation and the Synthetic Minority Over-sampling Technique (SMOTE) are applied to balance and expand the training dataset. The evaluation results demonstrate that the proposed TF- IDF and MLP model achieved excellent performance with an accuracy of 99.82%. The confusion matrix analysis indicates that precision, recall, and F1-score values were all equal to 1.00 across disease categories. Furthermore, the trained model was successfully converted into TensorFlow Lite (tflite) format for mobile deployment. Finally, the system was integrated into an Android-based mobile application named JagungKu to provide real-time diagnostic results. In conclusion, this research advances intelligent text-based disease detection systems and demonstrates the potential of deep learning in supporting sustainable precision agriculture.