Mitigating Algorithmic Bias in Multiclass CNN Classifications Using Causal Modeling

TL;DR

This work tackles gender bias in a multiclass CNN used for emotion classification by applying causal modeling in a post-processing framework. The authors train a CNN on a FairFace-derived dataset with DeepFace emotion labels for seven classes, then calibrate probabilities with a one-vs-all approach and fit per-class causal models to estimate gender effects. Debiasing adjusts predicted probabilities to reduce gender disparities while preserving overall accuracy, with test accuracy rising slightly from 60.4% to 60.8% and gender gaps narrowing across emotions. The study demonstrates that fairness improvements can accompany high performance in multiclass CNNs and extends causal-bias mitigation to deep learning, providing publicly available data and code for reproducibility.

Abstract

This study describes a procedure for applying causal modeling to detect and mitigate algorithmic bias in a multiclass classification problem. The dataset was derived from the FairFace dataset, supplemented with emotional labels generated by the DeepFace pre-trained model. A custom Convolutional Neural Network (CNN) was developed, consisting of four convolutional blocks, followed by fully connected layers and dropout layers to mitigate overfitting. Gender bias was identified in the CNN model's classifications: Females were more likely to be classified as "happy" or "sad," while males were more likely to be classified as "neutral." To address this, the one-vs-all (OvA) technique was applied. A causal model was constructed for each emotion class to adjust the CNN model's predicted class probabilities. The adjusted probabilities for the various classes were then aggregated by selecting the class with the highest probability. The resulting debiased classifications demonstrated enhanced gender fairness across all classes, with negligible impact--or even a slight improvement--on overall accuracy. This study highlights that algorithmic fairness and accuracy are not necessarily trade-offs. All data and code for this study are publicly available for download.

Figures (6)

• Figure 1: A Sample of the Labelled Data.
• Figure 2: Gender Fairness of the CNN Model (CM Training).
• Figure 3: Gender Fairness of the CNN Model (CM Test).
• Figure 4: Causal Model for Binary Classification. b5
• Figure 5: Gender Fairness of the CNN Model (CM Training, Debiased).