Scholarlink

The selection of crops in modern times should be done optimally considering various factors such as climatic conditions, soil composition, and pH; therefore, crop recommendation systems are a prerequisite for precision agriculture. This research explores machine learning, deep learning, and reinforcement learning models to predict crop selection based on the dataset containing 2200 samples, with features including macronutrient levels, pH, rainfall, temperature, and humidity, covering 22 crop types. The comparison involves the implementation of machine learning algorithms XGBoost, LightGBM, CatBoost, Quadratic Discriminant Analysis (QDA), and Explainable Boosting Machines (EBM) alongside deep learning methodologies, comprising Convolutional Neural Networks (CNN), Autoencoders, SAINT, TabNet, CNN BiLSTM with Bayesian Optimization and reinforcement learning algorithm (DQN). All these models are benchmarked against key metrics such as accuracy, precision, recall, and F1 score. After evaluating all the performance metrics, the most optimal algorithm, TabNet, was finalized. This will help the farmers, as by entering the macronutrients, pH, and weather condition data in the web application model, it can predict the best crop for the entered data. By this implementation, farmers could potentially maximize crop yield outcomes.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring Machine Learning Versus Deep Learning Versus Reinforcement Learning in Crop Recommendation Systems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Allu Pranav Srihas,
  • Abhijai Sasikumar,
  • K. Manikandan

摘要

The selection of crops in modern times should be done optimally considering various factors such as climatic conditions, soil composition, and pH; therefore, crop recommendation systems are a prerequisite for precision agriculture. This research explores machine learning, deep learning, and reinforcement learning models to predict crop selection based on the dataset containing 2200 samples, with features including macronutrient levels, pH, rainfall, temperature, and humidity, covering 22 crop types. The comparison involves the implementation of machine learning algorithms XGBoost, LightGBM, CatBoost, Quadratic Discriminant Analysis (QDA), and Explainable Boosting Machines (EBM) alongside deep learning methodologies, comprising Convolutional Neural Networks (CNN), Autoencoders, SAINT, TabNet, CNN BiLSTM with Bayesian Optimization and reinforcement learning algorithm (DQN). All these models are benchmarked against key metrics such as accuracy, precision, recall, and F1 score. After evaluating all the performance metrics, the most optimal algorithm, TabNet, was finalized. This will help the farmers, as by entering the macronutrients, pH, and weather condition data in the web application model, it can predict the best crop for the entered data. By this implementation, farmers could potentially maximize crop yield outcomes.