Salvaging this from another post... feel free to comment...

I have taken the full history stats and combined them into an auto-ranging weighted matrix resulting in just ONE pick per draw for 2 different jackpot games... Haven't hit much yet (a few $2 and $4 on the match6 so far, thanks to the 2 free QP's and a number or 2 from my line) but I finally did it... the pick changes every draw... follows the short term trends and is still rooted in the full draw history for it's root "profile"... and with excel it's all automatic ;-). While still no big bucks ,my picks are closer than when I was playing 5 and ten tickets per draw... Until I can find something new from calculus to apply, I will stick with this new "one shot" system for the forseeable future...

The math is sketchy, because it incorporates weighted averages, and backtesting only had a dozen $7 winners and 1 $100 hit, but based on all the work I have done looking for the "magic normulae" only to end up disappointed, I can play only one line per game knowing that I gave it my best shot...

the whole auto-ranging weighed average idea came from the study of neurons in artificial neural networks, with a component I added that gives more weight to the better pernorming numbers (using a simple match count as a weight multiplier). My "system" has an open architecture, allowing easy addition, removal or editing of individual components... right now I have 8...

Long Term Components...
MODE (full history) Provides best pernorming numbers
MEDIAN (full History) Provides a range
AVERAGE (Full History) smooths spikes
FORECAST (Full History) a "guess" component

Short Term Trend Components...
MODE (last 10 draws or less) provides recent "hot" numbers
MEDIAN (last 3 draws) short term ranging
AVERAGE (last 3 draws) another smoothing component
FORECAST (last 3 draws) a "wildcard" to shake things up, sometimes calls for negative numbers, but thanks to excel's ABSolute value function, it is now practical....

for each of the above components, a mode-based error correction mask was developed to maximize hits by position... this mask is applied and again checked against history to test and again mode-optimized by position...the resultant stat+error correction mask becomes one component (there are the above 8). A count of matches in the processed pick versus the actual draw numbers for the entire history is there to provide a hit count, this becomes the weight matrix... a component that has matched 19 hits in 179 draws gets a higher weight than a component thatonly matched 10, yet neither is discounted.

once they are all together, they are averaged by using the pick number x weight, this is done for each position to end up with one pick, which is then rounded to zero decimal places (instead of using INT wich discounts fractional components).

here's a simplified math example of the system in action using 2 components...

A. MODE of first position over full history = 2, EC mask = -1, comp number = 1
B. AVERAGE of last 3 draws = 9 Error correction mask = -2, composite number = 7

A has matched 17 of 179 draws, weight component (Awc) is 17
B has matched 12 of 179 draws, weight component (Bwc) is 12

normula = ((A*Awc)+(B*Bwc))/(Awc+Bwc)
result is 3.48, which rounds to 3 for your first position pick...