Practical Predictive Analytics

Time series models

Problems containing data observations which are not independent of one another, such as time-based data, deserve special treatment since events which occur in time exhibit a property known as autocorrelation. What that means is that the independent variables are all dependent upon the previous values of that variable. Many data techniques assume that all values occur independently of each other, so special techniques were developed to deal with time-based data.

Exponential smoothing is based upon the simple concept that the most recent data is the best predictor of the future. Exponential smoothing techniques allow you to tweak the parameters so that you can decide how much weight is given to the recent data versus the ...