Predicting events from set of Time Series

Question

I just want to get a general direction for my problem, as I have almost no experience with data science. I want to predict an event based on multiple trajectories.

E.g., for two given trajectories, a collision is imminent. So the input would be the two trajectories, and the label for these trajectories would be "dangerous". Similarly, any other pair of trajectories would be "not dangerous". A trajectory is a set of points, where each point is an x,y coordinate and may contain other information such as lane, size, etc. Of course, the order of points is important as this is a series over time.

I can freely generate trajectories and their labels for training purposes. At the end, I want a classifier that can classify trajectories as dangerous or not dangerous.

Question is, how should I proceed?

Answer 1

Logistic regression is a good place to start.

I'm guessing you'd get better accuracy if you transform your vector of points to a single trajectory value, like degrees or something similar. This could be done with linear regression or just some math. Simpler predictors are better, assuming you don't throw out too much information in the simplification process.

Medium is a good place for walkthroughs.