An experimental analysis of machine learning techniques for crop recommendation

Abstract

Taking a country into consideration where agriculture remains the primary occupation and farming still happens using conventional methods, the farmers are not able to produce anticipated yields. Modern farming strategies called precision farming play a vital role in improving crop yield and generating more profit for the farmers. This includes recommendations of crops that are suitable for specific fields based on soil conditions, temperature, rainfall, and humidity. To solve this problem, crop recommendation systems play an important role. In this research work, a crop recommendation system (CRS) was implemented using various machine learning algorithms that include random forest, decision trees, extreme gradient boosting (XG boost), and K-nearest neighbors (KNN). Experimental analysis was performed on the dataset collected from Kaggle. The Random Forest algorithm outperforms XG Boost, Decision Tree, and KNN with high accuracy and F1 score of 99.3% and 99.01% respectively. Hyperparameter tuning is additionally performed on XG Boost and Random Forest algorithms to improve accuracy. After hyperparameter tuning, the Random Forest algorithm outperforms XG Boost with an accuracy of 99.5%.