This page is for those interested in forecasting, time series analysis, data mining, predictive analytics and pattern recognition. I developed an interesting method, published in one of my papers. The method is based on a pattern recognition algorithm that is applied exclusively to time series data. I put together a simple Excel app written in VBA. Here is a brief intro:
Various methods are used to extrapolate time series and forecast the future values of an observed variable. Most of them are based on very rigorous assumptions that cannot be violated and they require users to be familiar with them. The approach used here is completely different. It falls more into a pattern recognition category, or you could even call it a data mining approach. I named the method APRE (Algorithm for Pattern Recognition and Extrapolation).
APRE searches for patterns in time series through a cluster of unobserved and artificially created vectors. Once patterns are matched, the distances between them are measured and they are used for fitting and predicting the future values of the time series.
In this current form, the method is implemented to be particularly suitable for stationary data sets.
To illustrate this approach, a simple Excel programme containing macros, was written. At present the first 40 observations of the famous Lorenz attractor are entered in the calls B5:B44. To predict the future values, just click, inside the spreadsheet, on the button with the time series picture on it, like this one: