2
$\begingroup$

So I have a dataset of trajectories of multiple devices. Each device has trajectory IDs (multiple trajectories), where each trajectory consists of start point (x, y) and end point (x, y), start time, and end time. Basically, each device is said to stop when it doesn't move for some time. When they move again, it's another trajectory.

The task is to predict the END end point (the x and y for the last trajectory) for each device, so it's a multi-label regression problem. The problem is each device will travel inherently different path, so I think I have to train a model for each device. I don't think it'll work since I'll have to predict entirely different set of devices, but let's put that aside for a moment. The main problem I'm having is that in the training data, some devices only have ONE trajectory and almost half of the devices have a count of 5 trajectories, so I think my solution is out.

How would you suggest to approach this problem using classical machine learning, assuming it's possible? If it's not, what NN architecture(s) should I consider? I'm not yet familiar with NNs but if it's the only way to do it, I'll try to quickly learn about the necessary architectures.

$\endgroup$
1
$\begingroup$

Any machine learning algorithm could possibly learn a solution once the problem properly defined.

One technique that could aid in learning is to bin the locations. Instead of predicting a specific x and y (and being precisely wrong), predict a more general region. Based on the topological structure of the location, bin similar location points together.

One class of classic machine learning techniques to try would be tree-based, such as decision tree, random forest, or boosted trees.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.