TIme Series Prediction Comparison with Linear, HMM, and SVM Models

by Theodore Van Rooy, Michael C. Mozer (Editor), Jem Corcoran (Editor)

A technique combining the intuition of Markov Chains and the classification power of Support Vector machines will be explored on its effectiveness at predicting highly stochastic time series. The time series will each be ana- lyzed with a basic Markov Model, and a Support Vector Machine algorithm. In all cases we are merely trying to predict whether the time series will have a positive or negative change in the next time step. Three sets of data are examined including Google and Apple Inc. stock price, Wisconsin dairy and cow production, and Seattle monthly rain and temperature levels. The results show that though the stock market and weather are still difficult to predict, the dairy industry can be predicted with a much greater degree of certainty. Furthermore the results indicate that with further research the financial industries and weather forecasters could viably include Support Vector Machines in their arsenal.

28 pages, Kindle Edition

First published December 1, 2006