How I Built an Elo Rating System for Esports Teams

#algorithms #esports #typescript #datascience

Why Elo for Esports?

Traditional esports rankings (HLTV, VLR) use a mix of editorial judgment and point-based systems. I wanted something purely mathematical — a system where every match result automatically updates team rankings.

The Algorithm

The base Elo formula is simple:

New Rating = Old Rating + K * (Actual - Expected)

Where:

K-factor = how much each match matters (I use 32 for regular matches, 48 for playoffs, 64 for grand finals)
Expected = probability of winning based on rating difference
Actual = 1 for win, 0 for loss

My Modifications

Standard Elo doesn't account for esports-specific factors, so I added:

1. Tournament Tier Weighting

const tierMultiplier = {
  S: 1.5,  // Majors, Worlds, TI
  A: 1.2,  // ESL Pro League, Masters
  B: 1.0,  // Regular tournaments
  C: 0.7,  // Qualifiers
}

2. Recency Decay

Matches from 6+ months ago contribute less to current rating. This prevents teams that were good a year ago but haven't played recently from staying ranked high.

3. Score Margin

A 2-0 sweep is weighted more than a 2-1 win. Close series suggest teams are evenly matched.

Results

After processing 50,000+ matches across CS2, Valorant, LoL, and Dota 2, the rankings at esport.is/rankings closely match expert consensus while being fully automated.

The system correctly predicted ~62% of match outcomes in backtesting — better than random (50%) and comparable to betting market implied probabilities.