Project · Rating & Clustering

Rating Demo + Player Cluster Demo

End-to-end workflow for player performance rating and role clustering: filter data, compare outputs, and produce delivery-ready results.

Open Demo API Docs

2Core modules

62Rating features

9+Target models

0.96+Avg R² score

Rating Demo

Rating Demo Capabilities

FutrixMetrics rating data is objective and model-driven, helping scouting and analysis teams evaluate full player performance quickly.

Rating Data Filtering and Comparison Layer

Use feature filters to isolate relevant player performance indicators before scoring. This keeps every discussion on the same sample base for consistent conclusions.

✓Apply precise filters across shots, goals, season, league, team, and core player indicators.
✓Optimize low-density metrics such as conversion rate and attacking involvement contribution.
✓Support multi-season lateral comparison for players in the same position.

FiltersCompare

▥

Rating DashboardFilter, rank, and compare model ratings across the dataset.

All positions Shots Goals 2024 / 25 Top 5 leagues

Player	Pos	Rating	Shot	Pass
Erling Haaland	ST	9.1	9.4	6.8
Jude Bellingham	CM	8.8	8.2	8.9
Bukayo Saka	RW	8.6	8.1	8.4
Josh Maja	ST	8.1	8.3	6.9

Rating Explanation and Delivery Layer

Rating outputs are designed for analyst review, report production, and client-facing delivery without repetitive manual formatting.

✓Return rating result objects that can be consumed directly by report modules.
✓Generate HTML player reports through the /report route.
✓Link outputs with model chart and player chart pages for unified presentation.

ReportsHTML export

Josh MajaPlayer report · ID 18244

Age 24ForwardFulhamPremier League2020/2021

Overall rating Weighted score from pro_rating.db · Per 90

Development profile

Attack

5.24

Creation

3.76

Passing

5.94

Progression

5.00

Defense

5.05

Discipline

6.98

GK Shotstop

0.00

GK Command

0.00

Appearance

2.52

Primary strengthDiscipline: foul / card control

Main riskGK command score missing

Best usageUse in the role's natural zone

Cluster Demo

Player Cluster Demo Capabilities

Built for three core tasks: role identification, similar-player retrieval, and structure analysis.

Cluster Role Identification Layer

First identify which role archetype a player matches, then evaluate fit against the current tactical system.

✓Search by player_name, season, club, and position.
✓Reverse-search by cluster_id, cluster_group, and cluster_name.
✓Useful for similar-player discovery and replacement evaluation.

SearchReverse lookup

▦

Cluster Feature DistributionFeature contribution by cluster group across role dimensions.

Player count by position

Forward

Midfielder

Defender

Goalkeeper

Roaming PlaymakerBall-winning MidNo-nonsense CBShadow StrikerWing-backTarget ManOthers

Cluster Structure Analysis Layer

Understand role-cluster distribution to assess squad balance more clearly.

✓Focus on cluster_group, cluster_local, and cluster_name dimensions.
✓Evaluate role distribution patterns across leagues or custom samples.
✓Combine with rating results to measure role-capability alignment.

LeaguesAlignment

◑

Cluster Structure SummaryRole distribution across leagues, aligned to rating results.

Defenders Forwards

No-nonsense Center-back (id=3)

9,226

False 9 (id=18)

5,479

Wing-back (id=1)

4,797

Shadow Striker (id=16)

2,623

Target Man (id=14)

1,475

Ball-playing Defender (id=5)

1,071

Full-back (id=6)

908

Libero / Sweeper (id=7)

666

No-nonsense Full-back (id=12)

652

Inverted Full-back (id=2)

629

Cluster size (log scale) · Top 10 of 20

Model Effect

Model Performance Overview

This section summarizes model design logic and validation outcomes for fast assessment.

“Quantified data makes player evaluation more objective. Even elite players vary by match and system, so data helps identify their best role, style, and strongest performance patterns.”

◎

R² by targetValidation score

Shooting

0.97

Passing

0.96

Defense

0.95

Aerial

0.94

Goalkeeper

0.89

▲

Baseline vs finalPer-metric lift

Rating

Assist

Foul

Baseline Final

⌁

RMSE by targetLower is better

Rating

Assist

Foul

Pass

Overall R² mostly above 0.96Low RMSE on rating & assistGoalkeeper shows higher error

Model Design Principles

Model by objective instead of forcing one model for every task
We train separate targets such as shooting_quality, passing, defense, aerial, and goalkeeper so each position and style is evaluated more fairly.
Shared core features plus target-specific signals
Most targets use 6–14 features, while rating uses more (62), balancing broad coverage with detailed judgment.
Validation performance is prioritized over training performance
Training and validation are evaluated separately to avoid overfitting. We prioritize Valid RMSE and R² for practical reliability.

Result Interpretation

Overall R² is mostly above 0.96, indicating model decisions are highly aligned with real performance trends.
Most targets show low RMSE, which means error is controlled for metrics like rating, assist, and foul_card.
Goalkeeper has a higher error level, indicating that keeper scenarios are harder and require more training samples.
Compared with previous versions, final outputs are generally better (lower RMSE), confirming that the upgrade is effective.

Output Examples

Random samples extracted from the rating and cluster databases. Browse, sort, filter, and export below.

2 Datasets

50 Rows each

13 Apr 2026 Updated

100 Total rows

Click any column to sort · Exports use the current tab and active filters

✦ Rating + Cluster demo

Explore the rating & cluster demo

Filter ratings, retrieve similar players by role, and export delivery-ready results — all from one demo workspace.

Open Demo ▣ API Docs