Crop prediction in Tamil Nadu according to environmental and soil factors using hybrid machine learning architecture

Abstract

Mathuranthagam, Tamil Nadu, India is the site of this research initiative that employs state-of-the-art hybrid machine learning (ML) architectures to forecast crop suitability in relation to environmental and soil characteristics. The model takes advantage of the strengths of linear support vector machine (SVM) classifier, bidirectional long short-term memory (BiLSTM), and convolutional LSTM (ConvLSTM) networks, and the data to capture complicated temporal and spatial correlations. To prepare the dataset for model training, it is normalized using min-max scaling and then feature selected using a Jaya optimization technique. The dataset contains variables such as humidity, rainfall, temperature, and pH. Both the BiLSTM and the ConvLSTM improve the model's comprehension of context from both previous and subsequent time steps. The ConvLSTM also records spatial dependencies. A powerful decision-making tool for differentiating across crop varieties is the linear SVM classifier. Comparing the hybrid model's performance to that of traditional LSTM approaches using measures such as recall, accuracy, precision, and F1-score shows that it performs much better. Using this approach can see how deep learning (DL) can supplement more conventional ML methods and see how important local environmental data is for agricultural policy and planning.