Hello,
I thought I'd like to share a method I've been working on which other people might like to use or develop further themselves. It uses recurrence analysis, a non-linear mathematical method developed to analyse noisy or chaotic time-series data. It has been successfully used to study a wide range of real-world phenomena, from the stock market to cardiac arrhythmias.
It occurred to me that it might be possible to analyse and predict lottery numbers using recurrence analysis. The mathematics is quite deep, but there is an excellent program I use which does all the hard work. It's called Visual Recurrence Analysis, and the current version can be downloaded from:
http://home.netcom.com/~eugenek/download.htm
Data files used are in plain ASCII format. Note this is the shareware version, so the 'save data' function is disabled, but you can save the graphs generated and study them using MSPaint (it's all you need).
I live in the UK, so naturally I am working with the results of the UK National Lottery. The game I am currently analysing is the Thunderball draw, which uses 5 main numbers drawn from 34, and the single 'Thunderball' drawn from 14. The recurrence model parameters I am using are:
Predictor: Nearest Neighbour
Distance: Euclidean
State Space Dimension: A range of values, 10 to 20
State Space Delay: Two values, 4 and 5
I am presently trying to refine the model, namely reduce the range of Dimension values, but even with the current values I can predict two to three of the main numbers per draw. I also hope to increase the number predicted, from 2 - 3 to 3 - 4. I think 4 accurately-predicted numbers is a realistic limit.
Why does it appear to work? I think the reason is the lottery machine itself. Over time, the physical parameters of the draw machine - its shape, size, number of rotating paddles, etc - determines how the balls move, ultimately imposing a degree of predictability on them. The patterns in the numbers reflect the underlying physics of what's going on inside the draw machine. A different design of lottery machine would impose a different pattern, but one which would still be amenable to recurrence analysis (you would simply need a different set of recurrence model values). The fact that the UK lottery employs more than one machine is not really a problem. They are all the same design, and so behave in the same way. It's effectively like using a single machine.
What originally led me to think lottery numbers could even be predicted? The book 'The Eudaemonic Pie' (also published as 'The Newtonian Casino') recounts the true story of how a group of American physics students developed a set of Newtonian equations to predict the path of a ball on a roulette wheel. The equations worked with up to 44% accuracy. Of course, there is a higher degree of randomness involved in the lottery (more balls, more complex behaviour), but I believe enough predictability is imposed by the draw machines to make prediction possible.
There are some questions I have yet to answer satisfactorily. Does analysing results by row or by column give better predictions? Similarly, should lottery results be analysed in numerical order, or by the order in which the balls were actually drawn? (at the moment I am using numerical order). Is there a theoretical limit to the number of balls which can be accurately predicted? (I said earlier that 4 might be a practical limit).
I am not claiming recurrence analysis is the perfect method for lottery prediction, but I do consider it to be the best (the only people who seem to get rich from commercial lottery predictor programs are the people who write them), simply because it is tailor-made for non-linear dynamical systems.