I am working on building a model to classify the type of touch the user makes (Long Press, Left Swipe, Right swipe, and so on). I have data with features that characterize the user's touch, like duration, velocity in x-direction, velocity in y-direction, etc. One feature that is also present is the trajectory of the touch.
The problem is that for touches like taps or long-press, the length of the trajectory array is 2 or 3 points, but for swipes, it reaches up to 40-100 points. What I thought could work is either using padding or CNNs. But the problem with padding is that as I am using trajectories, if I pad them with 0s it might affect the learning because a '0' still has meaning in trajectories as some points. And what I think the problem might be with CNNs is that first I do not know if such an architecture could work for all features (touch duration, xVelocity, etc.) as they are not spatially related. I may be wrong about this; feel free to correct it. I also thought of using RNNs, but I did not as they are mainly used for NLP tasks and all the features are not related to each other sequentially.
What are the different ways I can handle this kind of variable-sized input feature for neural networks?