AI Based Crop Recommendation Using Machine Learning Model

Small farmers often choose their crop, their fertilizer dose and their disease treatment by habit rather than by measurement. That costs them yield and money, and it puts extra chemicals into soil and water. This paper presents one tool that supports all three decisions. A crop model takes seven soil and weather readings (nitrogen, phosphorus, potassium, temperature, humidity, pH and rainfall) and was tested on 2,200 field records covering 22 crops. A 100-tree Random Forest and a Gaussian Naive Bayes classifier both reached 99.55% test accuracy, and rainfall and humidity turned out to be the most influential inputs. A fertilizer module then calculates how much Urea, DAP and MOP is needed to bring the soil up to the target of the chosen crop, counting the nitrogen that DAP already supplies, and adds a farmyard manure dose. A four-block convolutional network trained on 15,000 leaf images identifies common leaf diseases with 98.40% validation accuracy and returns a treatment suggestion. On a standard Intel Core i7 CPU a crop recommendation takes about 14 ms and a leaf diagnosis about 43 ms, so the whole system runs comfortably as a small web service.