Predicting Agricultural Crops from Soil Features in Chitradurga Area

Using soil parameters to predict crops can greatly improve farming results, as they are important factors that affect productivity. This study used machine learning models to predict which crops will grow best in the Chitradurga District based on the type of soil. Characteristics such as pH, macronutrients (N, P, K), and some micronutrients were examined using a large dataset of soil samples that were collected by hand from the Agriculture Department and APMC soil testing laboratories in six Chitradurga taluks. This public dataset, which is specific to this area, is the basis for a new crop prediction system based only on soil properties. Data preprocessing involved cleaning, normalizing, and addressing class imbalance using ADASYN and SMOTE. ANOVA F-score-based feature selection aimed to determine the most important soil characteristics. Four machine learning models, namely XGBoost, Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN), were examined to determine the best for predicting crop suitability. The XGBoost model achieved the best results, with an accuracy of 95%. The results show that soil characteristics can be used to make reliable crop recommendations and that data-driven methods can greatly improve decision-making in agriculture. This study shows that machine learning can be used in precision agriculture in Chitradurga, laying the groundwork for future improvements, such as hybrid modelling and the use of remote sensing data to make crop predictions that are even more accurate and specific to the region.