A PCA-SMO Based Hybrid Classification Model for Predictions in Precision Agriculture

Abstract

The human population is growing at an extremely rapid rate, the demand of food supplies for the survival and sustainability of life is a gleaming challenge. Each living being in the planet gets bestowed with the healthy food to remain active and healthy. Agriculture is a domain which is extremely important as it provides the fundamental resources for survival in terms of supplying food and thus the economy of the entire world is highly dependent on agricultural production. The agricultural production is often affected by various environmental and geographical factors which are difficult to avoid being part of nature. Thus, it requires proactive mitigation plans to reduce any detrimental effect caused by the imbalance of these factors. Precision agriculture is an approach that incorporates information technology in agriculture management, the needs of crops and farming fields are fulfilled to optimized crop health and resultant crop production. The proposed study involves an ambient intelligence-based implementation using machine learning to classify diseases in tomato plants based on the images of its leaf dataset. To analytically evaluate the performance of the framework, a publicly available plant-village dataset is used which is transformed to appropriate form using one-hot encoding technique to meet the needs of the machine learning algorithm. The transformed data is dimensionally reduced by Principal Component Analysis (PCA) technique and further the optimal parameters are selected using Spider Monkey Optimization (SMO) approach. The most relevant features as selected using the Hybrid PCA-SMO technique fed into a Deep Neural Networks (DNN) model to classify the tomato diseases. The optimal performance of the DNN model after implementing dimensionality reduction by Hybrid PCA-SMO technique reached at 99% accuracy was achieved in training and 94% accuracy was achieved after testing the model for 20 epochs. The proposed model is evaluated based on accuracy and loss rate metrics; it justifies the superiority of the approach.