{{ report_title }}

{{ report_subtitle }}

Model: {{ model_name }} ({{ model_type }}) Protected Attributes: {{ total_attributes }} Configuration: {{ config }}

Overall Fairness Score

{{ "%.3f"|format(overall_fairness_score) }}

{{ assessment }}

Overall Score {{ "%.3f"|format(overall_fairness_score) }}
Protected Attributes {{ total_attributes }}
Warnings {{ total_warnings }}
Critical Issues {{ total_critical }}
{% if total_critical > 0 or total_warnings > 0 %}

Issues Found

{% if critical_issues %}

Critical Issues ({{ total_critical }})

    {% for issue in critical_issues %}
  • {{ issue }}
    • {% endfor %}
{% endif %} {% if warnings %}

Warnings ({{ total_warnings }})

    {% for warning in warnings %}
  • {{ warning }}
    • {% endfor %}
{% endif %}
{% endif %}
{% if has_threshold_analysis %} {% endif %} {% if has_confusion_matrix %} {% endif %}

Fairness Overview

This report analyzes model fairness across protected attributes, measuring bias and discrimination in predictions.

Fairness Radar

Multi-dimensional fairness view across key metrics. Values closer to 1.0 indicate better fairness.

Fairness Metrics Explained

Statistical Parity

Measures if different groups receive positive predictions at equal rates. Value close to 0 indicates fairness.

Disparate Impact

Ratio of positive prediction rates between groups. EEOC requires ≥ 0.80 (80% rule) for compliance.

Equal Opportunity

Measures if True Positive Rates are equal across groups. Ensures equal benefit for qualified individuals.

Equalized Odds

Requires equal TPR and FPR across groups. Ensures both errors and successes are distributed fairly.

Treatment Equality

Ratio of FN to FP should be similar across groups. Ensures error types are balanced.

Metrics by Protected Attribute

{% for attr in protected_attributes %}

{{ attr.name|capitalize }}

{% if attr.pretrain_metrics %}

Pre-Training Metrics (Model-Independent)

{% for metric in attr.pretrain_metrics %}
{{ metric.name }}
{{ "%.4f"|format(metric.value) }}
{{ metric.interpretation }}
{% endfor %}
{% endif %} {% if attr.posttrain_metrics %}

Post-Training Metrics (Model-Dependent)

{% for metric in attr.posttrain_metrics %}
{{ metric.name }}
{{ "%.4f"|format(metric.value) }}
{{ metric.interpretation }}
{% endfor %}
{% endif %}
{% endfor %}
{% if has_threshold_analysis %}

Threshold Analysis

How classification threshold affects fairness metrics and model performance.

{% endif %} {% if has_confusion_matrix %}

Confusion Matrices by Group

Detailed breakdown of prediction outcomes for each demographic group.

{% endif %}

Generated by DeepBridge Fairness Suite

Report follows EEOC and ECOA fairness standards