How Our League of Legends Prediction Model Works

League of Legends is the most predictable of the games we cover — and our model leans into why: it reads the actual draft, not just the team names.

The single biggest factor: proficiency

The heaviest input (weight 0.24) is player-on-champion proficiency — each player's real history on the champion they've locked, role-weighted. A great player on a pocket pick is a different story than on their signature champion, and the model knows the difference.

Draft, objectives, and form

• Objective control — first dragon/herald/baron/tower and dragon-soul rates
• Champion synergy & lane counters — built from thousands of real pro co-pick pairs and lane matchups
• Player quality — gold/XP diff @15, damage share, role-weighted
• Team Elo, head-to-head, form, and patch-aware champion meta

We recently expanded the synergy and counter data to cover the great majority of champions actually picked in pro play — so off-meta and flex picks are now read more accurately than ever.

Measured, not marketed

On a leak-free, chronological backtest the model reaches roughly 66% on series winners — the strongest of our four games — and 70%+ on high-confidence picks across a large share of matches. Game-prop estimates (duration, total kills) carry honest error bars and are shown only where data supports them.

⚡ Build a draft and predict