Crop prediction

So basically my idea is like will use a rover kind of stuff which will consist of some sensors using which we will fetch values and store it on cloud and then later we can apply some machine learning algorithms based on which we can predict the crop. Here sensors are not yet fixed we are waiting for a dataset which we can get using which we can train the model. Currently we got npk value dataset but only npk value will not provide accurate results . So my question is how we can overcome this issue. Also if you guys have more ideas to add on it will be great.:grin: