This research paper introduces a machine learning-based approach for predicting crop yields using soil properties. The study focuses on four major crops: coconut, plantain, ginger, and tapioca, and examines key soil characteristics including pH, organic carbon, potassium, phosphorus, zinc, sulfur, boron, iron, manganese, and copper as classification features. The soil data used for analysis consists of 500 samples collected from the Agricultural Department of Kulashekharapuram Panchayat, Kollam District, Kerala, India. Three machine learning models, namely k-NN, SVM, and neural network, are implemented and compared in terms of accuracy and performance. The results demonstrate that the k-NN model outperforms the other two models, achieving an impressive accuracy of 88.8% using 10-fold cross-validation. This finding emphasizes the significant role of soil characteristics in predicting crop yields and highlights the efficacy of machine learning techniques in leveraging this information to enhance agricultural productivity. The proposed approach holds practical implications for farmers as it provides valuable insights for informed decision-making regarding crop selection and management practices. By utilizing soil variables, farmers can make optimized choices and improve their crop yield by maximizing the use of soil nutrients. This research contributes to the growing body of literature on the application of machine learning in agriculture, specifically in the domain of crop yield prediction. The findings offer valuable guidance for farmers, facilitating improved crop selection and ultimately contributing to enhanced food security.