Anyuhhh/pandaskill-lol-gold-data

PandaSkill Gold-Tier -- Match Dataset

This dataset contains approximately 8,768 Gold-tier ranked League of Legends matches collected from the NA server via the Riot Match v5 and Timeline APIs. It was used to train the role-specific XGBoost models available at Anyuhhh/pandaskill-lol-gold.

The collection and feature engineering pipeline follows the methodology of the PandaSkill framework (De Bois et al., 2025), adapted for amateur Gold-tier gameplay.

Files

File	Rows	Description
game_metadata.csv	~8,768	Match-level data: game ID, duration, patch, outcome
game_players_stats.csv	~87,680	Raw end-game statistics per player per match
game_features_15.csv	~87,680	Computed 15-feature vectors per player (paper feature set)
game_features_clean.csv	~87,680	Cleaned and filtered feature set used for model training
game_events.csv	~750,000+	Timeline events: champion kills, building kills, elite monster kills

Feature Description (game_features_15.csv)

Feature	Description
kill_life_assist_ratio	(kills + assists) / (deaths + 1)
gold_per_minute	Total gold earned / game duration in minutes
experience_per_minute	Total XP gained / game duration in minutes
creep_score_per_minute	(minions + neutral monsters killed) / duration
wards_placed_per_minute	Wards placed / game duration in minutes
damage_dealt_total_kills_ratio	Total damage to champions / total kills in game
damage_dealt_per_gold_total_kills	Damage dealt ratio normalized by gold obtained
damage_taken_total_kills_ratio	Total damage taken / total kills in game
damage_taken_per_gold_total_kills	Damage taken ratio normalized by gold obtained
largest_multi_kill	Maximum simultaneous kills in a short window
largest_killing_spree_total_kills	Longest kill streak normalized by total kills
worthless_death_ratio	Deaths with no team benefit in a 1-minute window
free_kill_ratio	Kills where the enemy death yielded no team benefit
objective_contest_winrate	Share of contested neutral objectives won
objective_contest_loserate	Share of contested neutral objectives lost

Data Collection Pipeline

Matches were collected using a Kafka producer-consumer architecture:

1. A Kafka producer queries the Riot Match v5 API for Gold I player match IDs
2. Match IDs are published to a Kafka topic
3. A Kafka consumer fetches full match data and timeline events, writing to PostgreSQL
4. PostgreSQL tables are exported to CSV for model training

Deduplication is handled at the database level using match ID as primary key.

Intended Uses and Limitations

Intended uses:

• Training player performance classification models for Gold-tier ranked play
• Research into extending professional esports analytics to amateur data
• Behavioral analysis and abnormal player detection in ranked games

Limitations:

• Data is limited to the NA region and Gold I rank tier
• Timeline events do not include damage statistics mid-game
• Match data reflects a specific meta period and may not generalize across patches

Citation

@article{debois2025pandaskill,
  title={PandaSkill - Player Performance and Skill Rating in Esports:
         Application to League of Legends},
  author={De Bois, Maxime and Parmentier, Flora and Puget, Raphael
          and Tanti, Matthew and Peltier, Jordan},
  journal={arXiv preprint arXiv:2501.10049},
  year={2025}
}