Agriculture is very important for food security, but over 40% of farmers use too little or too much fertilizer, which causes low crop yield, soil damage, and financial loss. Smart recommendation systems can help farmers by giving accurate advice on the right type and amount of fertilizer. Many machine learning methods like Decision Trees, Random Forest, Gradient Boosting, and Neural Networks have been used for crop and fertilizer recommendations. However, these methods often need a lot of computing power, do not work well with small or noisy data, and can be hard for farmers to understand. To solve these problems, we propose a Support Vector Machine (SVM)-based fertilizer recommendation model. SVM works well with small and unbalanced datasets, reduces overfitting, and handles complex patterns while needing fewer resources, making it suitable for real farming. Using soil nutrient values and crop needs, the model gives reliable predictions. Tests show that the SVM model achieves 96.77% accuracy, making it effective for smart agriculture and proper fertilizer use.