I have recently been working on using machine learning for yield prediction, however, I was exploring what inputs would be better at predicting yield. I am confused by only three papers that use historical yields as an input to predict yields for the new year. From the test results this does improve the prediction accuracy substantially. But does this count as data leakage? If not, what is the rationale for doing so? What are the limitations? (It seems that the three papers are from the same team.)
Three papers' links: https://www.sciencedirect.com/science/article/abs/pii/S0034425721001267
https://www.mdpi.com/2072-4292/12/8/1232
https://www.frontiersin.org/articles/10.3389/fpls.2019.00809/full
Pinery Lee Machine learning algorithms' predictions will assist farmers in deciding which crop to cultivate in order to maximize production by taking into account aspects such as temperature, rainfall, acreage, and so on.
Pinery Lee ,
You have 3 documents one which deals with wheat, corn and almond. My naive assumption without knowing the domain is that:
- Regional factors
- Type grain and sub type
- Sampling interval
Do matter.
One thing that I did notice is that they used publicly existing available data instead of having the sensors directly on the field. The limitation here is the level of granularity and assumption used. For example in [1] "averaging the Tmax and Tmin" on weather data without further support on why this can be done merits caution, since, averaging can distort underlying factors.
I suggest that you narrow down on a particular type of crop and look at these assumptions and their impact.
Regards
[1]https://www.frontiersin.org/articles/10.3389/fpls.2019.00809/full
You can. But it is not necessary to investigate how the point values of factors affect the point values of the results of cultivation. It is necessary to investigate how scenarios of changes in point values of factors affect scenarios of changes in point values of the results of their influence. And the created models will be adequate only during the duration of the ergodicity period. If you have such data, then we can conduct a joint study using the technology of scenario automated system-cognitive analysis created by me. I'm ready. The period of ergodicity is the period during which the patterns in the subject area do not change. Sanctions violate ergodicity. Now we are at the point of bifurcation, when new patterns are being formed.
If we analyze the prediction-related database, we can confirm that there are input variables and output variables. If all input variables are used, a problem arises that the processing time becomes longer. Therefore, it is important to select and use meaningful input variables.
Assuming that n input variables are used as a method of selecting meaningful input variables, the combination with the lowest error value can be selected.
Yes you can. Predictions are predominantly based on Historical values.
- Badges
- Science topic
- Similar topics
- Counseling Psychology
- Confusion