[Ostin]

I’ve been messing around with spreadsheets for years, trying to write my own little model for Premier League scorelines, but I keep bumping into the same wall: too many variables, not enough brain‑power. Last night I watched City ‑ Spurs and my “gut” said goals, yet my sheet spit out a cautious 1‑0. Of course it finished 3‑3 and I felt silly again. Has anyone here moved from DIY formulas to full‑blown AI tools? I’m curious whether neural nets really pick up patterns humans miss, or if that’s just marketing smoke. Would love to hear how people blend machine smarts with regular footy sense in practice

[Buddy Dorsey]

I used to run predictions off Elo ratings alone, and honestly it was hit‑or‑miss. The leap came when I started feeding match event streams (pressures, progressive passes, xThreat chains) into a small recurrent network. Accuracy nudged up, but more importantly my confidence intervals got tighter—I could finally tell when the model was truly unsure. If you want to shortcut months of Python tweaking, the guys over at Best Soccer Prediction Site in the World already bundle an ensemble of gradient‑boosted trees plus a lightweight LSTM layer. I subscribe mostly for inspiration: I’ll grab their probability range for, say, “Over 2.5”, compare it with my own output, and dig into mismatches. The cool part is the way their daily dashboard explains *why* the algorithm leans one way—positional adjustments after early‑week injuries, travel fatigue scores, that kind of thing. It’s not a magic oracle, but it’s miles beyond my lonely spreadsheet.

[Author]

Posts