Comprehensive Analytics
BiasClean v3.0 gives you a clear and simple way to understand fairness in your dataset. It turns complex statistics into easy-to-read charts, tables and explanations. There is no jargon, and you do not need any background in data science.
1. What “Comprehensive Analytics” Means
The analytics system checks your dataset from different angles and shows:
Where groups are under-represented
Where groups appear too often
How serious each imbalance is
How much it affects fairness in your chosen domain
How the system improves fairness after rebalancing
Everything is explained in clean, everyday language so anyone can follow it.
2. What You See in the Analytics Report
When you upload your data, the system creates:
A. Group Representation Table
Shows how many people belong to each demographic group (e.g., age, gender, ethnicity) and how this compares to what is expected in the real world.
B. Traffic-Light Colours
BiasClean uses simple colours:
Red – serious unfairness
Amber – medium unfairness
Green – balanced and healthy
The colours help you understand the problem at a glance.
C. Domain Awareness
The system adjusts results based on the domain you select.
For example:
Justice: Ethnicity and SES matter more
Health: Disability, Gender and Ethnicity matter more
Finance: SES and Region matter more
This makes the analytics smarter and more realistic.
D. Weighted Fairness Score
This is a number that shows how fair or unfair the dataset is overall, based on the seven UK fairness features and their domain weights.
A lower score means more unfairness.
A higher score means better balance.
E. Visual Charts
BiasClean shows clean charts such as:
Before vs After fairness
Group balance charts
Domain-weight impact charts
Data-retention charts
They help you “see” what the system is doing.
3. Before and After Comparison
One of the most helpful parts of the analytics page is the “Before vs After” section.
It shows:
How the dataset looked before rebalancing
How it looks after SMOTE and controlled undersampling
How fairness improved
How much data was kept
This helps users understand exactly what changed.
4. Why These Analytics Matter
Without fairness analytics:
A dataset may look fine at first glance
Hidden biases stay unnoticed
AI models may treat certain groups unfairly
Decisions can become unreliable or discriminatory
With BiasClean analytics:
Problems are visible
Improvements are measurable
Reports are easy to read and share
Fairness becomes transparent and explainable
These analytics are suitable for:
Students
Teachers
Researchers
Businesses
Regulators
Anyone who wants a fair dataset
5. Designed for Non-Technical Users
The system never expects you to know statistics or coding.
It explains results using simple words, such as:
“over-represented”
“under-represented”
“balanced”
“fairer than before”
It also provides short explanations next to charts so that anyone can understand the results immediately.
6. What You Can Download
After the analytics run, you can download:
The full analytics report (PDF/HTML)
The rebalanced dataset (CSV)
The summary of fairness scores
The visual charts of before/after results
This makes sharing with colleagues or regulators very easy.
📊 How to Read the Analytics Report
The analytics report in BiasClean v2.0 is designed to be simple, clear and beginner-friendly. You do not need any maths or coding knowledge. This guide explains how to read each part of the report step by step.
1. Start With the Overview Box
At the top of the report you will see a short summary showing:
Fairness Score (Before) – how fair the dataset was at the start
Fairness Score (After) – how fair it is after rebalancing
Data Retained – what percentage of the original data is kept
Rebalancing Mode – Standard or Industry Mode
This gives you a “big picture” view of how much improvement happened.
2. Look at the Group Representation Table
This table shows how many records belong to each group, such as:
Age bands
Gender
Ethnic groups
Regions
Disability status
Migration status
Each group has two values:
Expected % – based on real UK population patterns
Your Data % – what your dataset actually contains
If a group’s Your Data % is far from the Expected %, it means the dataset might be unfair.
3. Use the Traffic-Light Colours
Colours make it easy to spot problems:
Green – Good balance
Amber – Medium imbalance
Red – Strong imbalance
If you see many amber or red areas, it means the dataset needs attention.
4. Check the Domain Weights
When you pick a domain (for example, Justice or Health), the system uses different importance levels for each feature. This is called a domain-specific weight.
For example:
In Health, Disability and Gender matter a lot more.
In Finance, SES and Region matter the most.
These weights help the system decide which imbalances are the most serious.
5. Read the Disparity Explanations
Next to each major finding, you will see a short explanation written in plain English.
This answers questions like:
Which group is under-represented?
How much is the gap?
Why does it matter in this domain?
Is the difference large enough to cause unfair outcomes?
These explanations connect the numbers to real-world meaning.
6. Compare “Before” and “After” Charts
The report includes visual charts that show:
How groups were distributed before rebalancing
How they look after SMOTE and controlled adjustments
How the fairness score improved
How much data was added or removed
These charts help you see whether the dataset is now more balanced.
7. Look at the “Production-Ready Status”
At the end of the report, the system shows a final message:
Production-Ready
Needs Review
Or Diagnostic Only
This tells you whether the new dataset is suitable for real use in:
AI models
School or university projects
Business decisions
Research
Compliance checks
8. Download Your Files
When you finish reading the report, you can download:
The rebalanced dataset (CSV)
The full analytics report (PDF/HTML)
Visual charts
A summary sheet for sharing
These files are easy to give to teachers, colleagues, regulators or your technical team.