Authors - Aishwarya V. Kadu, KTV Reddy Abstract - This research investigates the effectiveness of machine learning and deep learning models in multi-class crop classification using an agricultural dataset comprising 3,450 records from the Wardha district of Maharashtra, India. The dataset includes essential soil and climatic parameters such as pH, nitrogen (N), phosphorus (P), potassium (K), Mn, Fe, Cu, Zn, S i.e Nutrients and micronutrients temperature, humidity, wind speed, Cloud and rainfall, covering 11 crop types. Data preprocessing involved label encoding for su-pervised learning tasks. The study developed and evaluated three predictive models: Random Forest, XGBoost, and Multi-Layer Perceptron (MLP). Model performance was assessed using classification metrics like accuracy and confusion matrix, as well as regression metrics such as MAE, MSE, RMSE, and R-squared. Feature importance analysis was performed to deter-mine the influence of input variables on model predictions. XGBoost showed the highest classification accuracy, while MLP demonstrated strong learning capability. The findings emphasize the value of combining predictive performance and interpretability, highlighting the applicability of AI-driven methods in enhancing crop recommendation systems and advancing precision agriculture across varied agro climatic zones.
