0

I'm just starting in machine learning learning models such as line, ridge regression, perceptron, logistic regression,gradient descent and i don't see how any of these models could be used to tackle this modelling problem ? Can anyone provide me with an hint on what to use ? Due to the periodic nature of the problem i'm inclined to search time series, however this seems like an overkill cause that subject hasn't been explored yet in my class.

For the features i would be inclined to check the tides size, coastal length, and moon phase. I also think, since the data is granularity is at most hourly, we should learn daily instead of monthly (The data size isn't too overwhelming) and we could gather a better insight.

Consider the problem of predicting the tides of a given location considering the daily and monthly cycles. Assume also that you have access to hourly data from a period of 10 years. Define and discuss how would you model this problem: What features would be considered? How could the system learn? Should we learn the daily and the monthly cycles separately?

John Doe
  • 59
  • 7

2 Answers2

0

What do you want to "learn" ? I think you can use some distance function and try some statistical classification like K-Means and then predict using KNN on your clusters

A K
  • 41
  • 3
0

Do you want to predict the tide(high/low) based on time of day at a location?

I would first use features like moon phase(you can define a feature vector with one hot encoding for every phase), time of day feature( again one hot encoding for each hour(or whatever discretization you want)); and maybe length of coast ; Concatenate all these features(phase, time of day, coastline len) to make a Feature vector

Now if you want to predict high/low tide using your dataset: since this is periodic in time I would use a complex non linear model like a Random Forest or Maybe Gradient Boosted Decision Tree which would have the power to learn this periodicity.

Abhimanyu
  • 134
  • 8