This is an interesting problem. You can use the regression techniques discussed in the book or the classification techniques. You have to first determine carefully what you want to predict, and what is the information you will use to predict it. You should use all information available to you about the nature of the problem when doing so. Here are two suggestions.
1. Predict the average temperature for the next month given all prior history. This is a regression problem. You might choose as your inputs to your regression the average temperature of the current month; and the historical average temperature for the month that you are trying to predict. The target is ofcourse the next months temperature. If you choose your inputs carefully, a linear regression might suffice.
2. Predict whether the average temperature for the next month will be above or below its historical average. This is a classification problem. You could use the same inputs as in 1, and if the inputs are chosen carefully enough, a linear classification might suffice.
So depending on how you define your problem it is classification or regression. I suspect the success of your solution will depend heavily on how you choose your inputs on which to base the prediction (using as much information about the problem as possible before looking at the data, such as the 12month periodicity of weather, etc.)
Quote:
Originally Posted by invis
I have a dataset with ~1500 data points that looks like:
1922 / 2 / 3
year / month / average temperature.
And I want to make a predict the temperature by year and month. What the technique is the better to choose here ? Almost all what we learn in this class is about classification, not to predict, so I am confusing a bit.
