Lab
The method
For every Loto draw since 6 October 2008 (Super Loto and Grand Loto included), each strategy generates a grid knowing only the previous draws. That grid is then compared with the real result, counting matched numbers, matched Chance numbers and winning grids (using the current 9 prize tiers).
By pure chance, a grid matches 0.51 numbers and 0.10 Chance numbers on average, and wins a prize about once in 6 draws. Change the seed to replay with different random draws: the ranking of the strategies changes every time, a sign that none of them has a real edge.
The experiment
Artificial intelligence versus chance
What if real artificial intelligence could find what simple statistics miss? We trained three machine-learning models on the full history. They learn draw after draw, never seeing the future, and each time play the 5 numbers and the Chance number they consider most likely.
- Logistic regression
- Weighs five signals for each number (frequency over 10, 50 and 200 draws, gap, drawn last time) and learns how much each one matters.
- Neural network
- The same signals go through a network with 8 hidden neurons, able to pick up non-linear combinations. Its starting weights depend on the seed.
- Markov chain
- Learns which numbers come up after which others: if 12 were often followed by 34, it would play 34 after every 12.