Using Machine Learning (Or Any other Technique) to recognize Patterns in a Graph?

I'm no AI expert but I'd say AI isn't required to solve this and most probably wouldn't be the quickest way. I would try this approach:

Imagine a cube - axis X is the day of the week, axis Y is the time spent studying and axis Z is the frequency of that happening (reflecting the "mostly" and "rarely" in your graph).

You implement the cube as a 3-dimensional array of Integers, the Y axis obviously having a size of 7 and for the X and Z axis you choose some size that seems reasonable to you (maybe increments of 30 minutes for X and frequencies of once a week, twice a week, ..., every day for Z).

Then you simply fill in your sample points into the cube by incrementing the Integers at the according indices by one (initially zero).

Finally you can find concentrations (maxima) of studying-activity by "moving" a "window" (a smaller cube) through that cube and summing up the Integers in that window. Example: Let's say your cube has the dimensions of (X,Y,Z) = 6,7,7 (6 steps of 30 minutes, Monday through Sunday, 1-7 times per week). The "window-cube" could have the dimensions of 2,2,2 and you move it through the cube within 3 nested for-loops, each increasing one of the coordinates, starting at 0,0,0. You sum up the values of the Integers in the cube that overlap with the window-cube (which will require some more nested for-loops) and store that value in some other structure together with the coordinates of the window-cube at that point. When you're done, you sort the results by the sum and you have your results. You could experiment with different window-sizes and -shapes (e.g. a window of 1,1,1 or 1,1,7 or 6,1,1 or what have you).

If you try this and it worked for you, please drop a comment, I'd be interested to hear about it :)