How we can do Sunspots predict for next 3 months or 6 months or 1 year in the future.
First you’d need a substantial dataset for sunspot information vs time.
Since you want to predict a time series, a good choice would be a Recurrent Neural Network, that’s the topic of Course 5.
Interesting question. Tom has described what you would need to approach this problem using a Deep Learning network. You need a lot of training data and then to choose an appropriate network architecture.
But maybe the higher level question to consider first is whether ML is the right approach to this problem. I have never really looked into sunspots, but I would think that this is more of a physics problem than a pattern recognition problem. The point is that Machine Learning can only recognize patterns that it learns from existing data. Someone asked a couple of years ago whether you could use Deep Learning to predict the stall speed of an airfoil in a wind tunnel. But this is just a aerodynamics problem and what if the new airfoil is a different shape than all the previous ones represented by your training data? The better strategy for that problem is to invest your effort to build a aerodynamics simulator that implements Bernoulli’s Law rather than gather a bunch of training data and trying to train a neural network.
But this is all with the disclaimer that I don’t know anything about what really causes sunspots, so maybe the “learn from previous data” approach actually has merit.
Thinking about it a bit more the other thing to understand would be what kind of data is available from astronomical observations of the sun. Maybe if you have images that are detailed enough that include infrared wavelengths or the like there might be patterns of temperature activity that are precursors to sunspots and you could train a model to recognize those patterns.