Tuesday, March 30, 2010

Using Singular Value Decomposition (SVD) to factorize data on car properties


This file has data on five cars: Chevrolet Matiz, Chevrolet Evanda, Chevrolet Tacuma, Chevrolet Nubira, Chevrolet Kalos. The properties considered are engine displacement, length, width, cargo volume, number of cylinders, power, fuel consumption and CO2 amount.
SVD reveals that 93,4% and 5,4% of information is contained in the first two singular vectors, respectively – leaving only 1,2% for the rest of the decomposition. The first singular vector can be called general assessment – it shows a basic relationship between all the cars across the entire set of variables. Our main focus is to try to understand the contribution that the second singular vector has. As shown above, the most influential contributions of the second singular vector are in engine displacement, cargo volume and power (kW) – this is the fine-tuning, which shows that the cars can’t simply be described with a rank one matrix.
This is the second file on this blog, that was not originally created by me (the first was the model of bird flocks), but rather replicated and expanded.

No comments: