Dimensionality reduction

Is there any clear separation between data points of one class vs. another class?

If dimensionality reduction solved the problem, it would be a good enough classification method in itself.

Goal of dimensionality reduction: take any data point that has a gazillion features/coordinates (temperature, pressure, location, …, user id, procrastination tendencies) and represent it with 2 or 3 coordinates (x,y,z) → plot on 2D plane or 3D space

If the distance between pairs of points in high-D space is the same as distance between pairs of points in low-D space, for small distances (i.e. points that are close to each other) that means local structure is preserved.

If this distance is preserved for large distances (points that are far apart), that means global structure is preserved.

Dim reduction could have different goals.

One goal is to visualize the data points and see if data points in the same are distributed in some neighborhood/cluster. If yes, a new data point can be classified by dim reduction.