Agricultural Crop Recommendation, Crop Disease Detection and Price Prediction Using Machine Learning

Abstract

India's foundation is its agriculture. With over 60% of the workforce employed and producing over 18% of the nation's GDP, it is a vital sector of the Indian economy. Although there are many ways in which we can use technology to increase product production, a farmer can only profit if he is able to sell his crops. Three laws have been passed by the Indian government to encourage the export of agricultural products across the nation. But today, we witness farmers all over the nation fighting against these regulations to protect their rights. Farmers worry that big merchants will exploit them as puppets and undercut the price at which they sell their goods. After doing a thorough analysis of the situation, we developed the concept of creating an agricultural produce application that facilitates direct communication between farmers and retailers, allows for product reviews and crop yielding rate prediction, and predicts the price of agricultural produce based on quantity produced and previous years' sales rates. Unpredictable rains, unexpected temperature decreases, and heat waves have all been brought on by the shifting climate, and the ecosystem has suffered significant harm. Thankfully, machine learning has produced useful methods for tackling international problems, such as agriculture. These climate change-related agricultural issues can be resolved by using various machine learning methods. The purpose of this piece is to Create a method to identify crop diseases and suggest crops. For both objectives, publicly accessible datasets were utilized. Regarding the crop recommendation system, feature extraction was done, and a variety of machine learning methods were used to train the dataset, including Support Vector Machine (SVM), Random Forest, Decision Tree, Logistic Regression, and Multilayer Perceptron. 99.30% accuracy was attained via the random forest algorithm.CNN architectures such as ResNet50, and EfficientNetV2 were trained and compared for the plant disease identification system. EfficientNetV2 outperformed the rest, with a high accuracy of 96.08%.