Nonlinear regression
17fev2015 - Euromillions
Data file processed with B-spline Order =3 Interval = 0.0125 (insert 80 intermediate datas between 2 numbers)
Predictor : Nearest neighbour
Distance : Euclidian
Dimension 3
Delay 2
Bandwidth 9
Data window used = 48000
17/02/2015 02 05 18 30 43 /1 10
Results at 80-160-240-320-400
Model Results
---------------------
Time Index Actual Value Predicted Value Error (NMSE)
48001 0.5440586 1
48002 -4.8893911 2
48003 6.3601678 3
--------------------------
48078 29.1009668 78
48079 29.5949999 79
48080 30.1223886 80----30
48081 30.6559556 81
48082 31.1638777 82
--------------------------
48155 43.0777338 155
48156 42.8645999 156
48157 42.6683116 157--43
48158 42.4350446 158
48159 42.1979336 159
48160 41.980545 160-------
48161 41.7508774 161
48162 41.5136002 162
--------------------------
48235 17.9162668 235
48236 17.5252889 236-18
48237 17.1408666 237
48238 16.705722 238
48239 16.2460999 239
48240 15.8173555 240--------
48241 15.4068667 241
48242 15.021411 242
------------------------
48317 1.8668622 317
48318 1.8733389 318
48319 1.8780378 319
48320 1.9465211 320-----2
48321 2.0723756 321
48322 2.1588112 322
--------------------------
48398 30.6559556 398
48399 31.1638777 399
48400 31.6756001 400---
Just an example.
That's a multi-step prediction, 400 datas ahead predicted with results at each 80 datas.
4 first numbers are good or near good. Fifth number is more difficult to find.
I generally use 2 numbers at each 80 datas. 3 good numbers on that drawing plus one star.
One-step prediction would be more accurate but the task is long and tedious.
I have a program beeing coded to automatize the process.
I have others examples like that.
I have won a lot with nonlinear regression. The only way to win.
Bob