aidancoit9203/ufc-round-stats

UFC Round Statistics

Round-by-round striking and grappling statistics from the UFC, one row per fighter per round, with the bout and event context joined on.

Generated from the Cito UFC API — a self-serve UFC and MMA data API. Regenerate any time with the script included in this repo; the corpora are derived from public fight data, not scraped by consumers.

What's inside

Config	Rows	Grain	What it is
round_stats	~2,300	fighter × round	Strikes, takedowns, control time with target and position splits
dwcs_prospects	~30	bout	Dana White's Contender Series bouts with both fighters' profiles

The round_stats set covers ~500 bouts and ~620 fighters across 40 recent events.

Why this dataset is useful

The upstream API returns strikes as combined strings ("3 of 5") and control time as a clock ("0:11"). This dataset splits those into clean integer columns and adds derived accuracy fields, so it loads straight into a dataframe with no parsing work:

• sig_strikes_landed, sig_strikes_attempted, sig_strike_accuracy
• takedowns_landed, takedowns_attempted, takedown_accuracy
• control_time_sec
• Target splits: head_*, body_*, leg_*
• Position splits: distance_*, clinch_*, ground_*

Schema — round_stats

Column	Type	Notes
event_slug, event_title, event_date	string	Event identity
bout_id	string	Join key to a bout
weight_class, is_title_bout, card_section	string/bool	Bout context
fighter_slug, fighter_name	string	The fighter this row describes
opponent_slug, opponent_name	string	The other corner
result	string	Finish method, e.g. "Submission", "KO/TKO"
result_round, result_time	int/string	When it ended
winner_slug	string	Winner's slug
round	int	Round number
knockdowns	int
sig_strikes_landed / _attempted / sig_strike_accuracy	int/float	Significant strikes
total_strikes_landed / _attempted	int	All strikes
takedowns_landed / _attempted / takedown_accuracy	int/float
submission_attempts, reversals	int
control_time_sec	int	Parsed from "M:SS"
head_landed / head_attempted	int	Target split
body_landed / body_attempted	int	Target split
leg_landed / leg_attempted	int	Target split
distance_landed / distance_attempted	int	Position split
clinch_landed / clinch_attempted	int	Position split
ground_landed / ground_attempted	int	Position split

Load it

from datasets import load_dataset

ds = load_dataset("aidancoit9203/ufc-round-stats", "round_stats", split="train")
print(ds[0])

prospects = load_dataset("aidancoit9203/ufc-round-stats", "dwcs_prospects", split="train")

import pandas as pd
df = pd.read_json("hf://datasets/aidancoit9203/ufc-round-stats/round_stats.jsonl", lines=True)

Example

Top strikers by significant strikes landed across the covered events:

import pandas as pd
df = pd.read_json("round_stats.jsonl", lines=True)

(df.groupby("fighter_slug")[["sig_strikes_landed", "takedowns_landed", "control_time_sec"]]
   .sum()
   .sort_values("sig_strikes_landed", ascending=False)
   .head(10))

Known limitations

• dwcs_prospects is sparse. Contender Series fighters are not yet on the main UFC roster, so *_country, *_division, *_age and *_stance are frequently null. Record and reach are reliable.
• opponent_name is null for ~2% of round rows where the corner could not be matched.
• sig_strike_accuracy and takedown_accuracy are null when attempts were zero (avoids divide-by-zero).
• Coverage reflects recent events, not the full historical catalog. Increase the sweep with --events N.

Regenerating

pip install httpx
export CITO_API_KEY=your-key       # free: https://citoapi.com/signup/
python generate.py --dataset all --events 40

Source: https://github.com/achillesscriptsvip/cito-ufc-kit/tree/master/datasets

Attribution

Data compiled by Cito API from publicly accessible sources. If you publish work derived from this dataset, please cite Cito API and link to https://citoapi.com/ufc-api/.

Free API key (500 calls/month, no credit card): https://citoapi.com/signup/

Licence and disclaimer

Released under MIT. This dataset is independent and is not affiliated with, endorsed by, or licensed by the UFC, Zuffa, TKO Group Holdings or any promotion. All trademarks belong to their respective owners.

Odds data, where present elsewhere in the API, is third-party market data for research, media and display — not for betting settlement. This dataset contains no odds.