1
$\begingroup$

I don't have a solid background in statistics. I am double checking with you on a phenomenon I am trying to study.

we are doing a study of some very rare species of flowers. We are putting them in different places in our farming area. We are trying to predict the time when butterflies are coming and when they are leaving. Butterflies are coming in specific time, they are feeding from these flowers, when butterflies come to place x in our farming area, they are looking for flower Y some of the flowers are already occupied, and some times, the whole flowers are all occupied, so they leave or go to other places looking for another flower in another place. We collected a dataset of Arrival/Departure of butterflies in the farming area. we want to analyze the dataset to be able to predict their next move. we have different locations (10-20 places).

Dataset:

  • Arrival time

  • Departure time

  • isThere noise: time when there is a noise, usually butterflies don't come when they hear noise.

Question:

We are trying to choose which model works best here.

Some points to double check with you. SHould we divide our dataset to different cells, and analyze every location separately.

If no, Some cells are spatially near to each others. Should we consider that. Again, which model should we use to predict if the Butterflies are coming and when they are leaving X (location based on our grid-division). we are looking for a good approach to build a predictive model to arrival-departure.

Note: For those who are wondering why we do that, it's very tedious problem to manage the presence of humans in farming areas. We don't want to scare butterflies, it's more rewarding for us to have them come to feed from flowers (there is a deep discussion here), so we need to manage the time we should be there.

Last thing, we are not researchers, we practitioners who want to use data/tech to manage their farming industry. I know that this is funny for some of you to see someone using ML to take care of his flowers, but this is cool

Improve this question
$\endgroup$
4
  • $\begingroup$ Sounds like an interesting question. By "the time when Butterflies are coming and ... leaving" do you mean time of day? (Some people studying pollinators are interested in how often they arrive and how long they spend at flowers, but it seems your focus is when during the day they show up). Regarding the fact that some of your flowers are near to each other, there are several ways to address this, including models with spatial autocorrelation and random effects models. What kind of software will you do the analysis in? $\endgroup$
    – N Brouwer
    Commented Feb 21, 2015 at 5:35
  • $\begingroup$ Exactly @NBrouwer I pointed that I am not a researcher, just to clear these doubts.I am interested in the probability of every flower occupation in specific hour/minute of the day. That would influence our production/ flowers' life. I can use Python (I have some experience, yet not solid, in building ml things). Can you throw some hints, and how can I handle this one. At the end of the day, these flowers are coming back home, they will not staying flower. I don't think that I can model that using Poisson probability distribution (number of butterflies in every flower in time (t)),then predict. $\endgroup$
    – user3378649
    Commented Feb 21, 2015 at 6:22
  • $\begingroup$ You sound like you have an interesting data set. You could always put it online and someone may analyse it for free. $\endgroup$
    – csgillespie
    Commented Feb 21, 2015 at 15:42
  • $\begingroup$ @csgillespie I'd rather learn how to fish, then eat the fish ;) But it's a good suggestion $\endgroup$
    – user3378649
    Commented Feb 21, 2015 at 23:56

1 Answer 1

Reset to default
0
$\begingroup$

One way would be to analyze this would be with a logistic regression, something like this

$$Pr(\text{Flower occupied}) = \text{time of day} + \text{noise?(Yes/No)} + \text{anything else}$$

or

$$ \text{Number of flowers occupied/number of flowers in group} = ...$$

A very flexible modeling approach that could accommodate non-linear changes in activity, as temperature increases and decrease throughout the day, is generalized additive model (GAM) which can fit various forms of smooth curves to data. They can accommodate modeling probabilities and binomial frequencies. I do not use Python but the NumPy or SciPy packages probably can do these. The R package mgcv is what I use. If you use standard logistic regression you could fit a polynomial term such as time$^2$ and time$^3$ but these have drawbacks and limitations.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks for ur answer. I'd give it shoot today and see the results. $\endgroup$
    – user3378649
    Commented Feb 21, 2015 at 23:58
  • $\begingroup$ I couldn't reach good resulst with regression. However, i am checking the possibility tp model this problem using Hidden Mokov and throwing the model on temporal (Poisson) and spatial information (Gaussian) distributions. $\endgroup$
    – user3378649
    Commented Feb 24, 2015 at 20:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.