I have a table with data about 10 agriculture parcels. Each parcel has data in time regard the number of nutrients each parcel has received in each day and in the end I have the total number of oranges that were harvested in the end of the season, so the headers look like this:

id     water_10102019   nitrogen_10102019 ....water_30072020   nitrogen_30072020  potassium_30072020   total_oranges

My task is to tell if it is possible to predict the number of oranges from this dataset, and also, to show some graphs/statistics for this. What I did at the beginning, not "by the book", was to just put each date in regression and see which date has the highest correlation with the total oranges in the end. I did it using multivariate regression, and I got for some dates a very high R2.

My problem is that = I did not expect the R2 to be so high just from running regression like this, per date, especially when it is known that predicting the number of oranges is very hard task. I saw that many of these dates have no normal distribution, high kurtosis ect. I feel like 10 parcels is not enough data.

However, I still have to produce something.I have generated graph with R2 during the time, but I feel like is really biased,as is only 10 data points , that doesn't have normal distribution at every checked date.

As I'm only starting my way in the data world, I would like to get advice- is my idea to do correlation this way makes sense? should I have other order of operations that you would recommend when approaching a task like this? something that you always do when you get this set of data? and what should I do if I think that this is not enough data but I still need to show the potential of predicting something?

Eager to learn from your ideas and advice:)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.