Mitigating bias in data analysis

Conquering Data Bias

Konstantinos Kattidis

Data Analytics Lead

Conquering cognitive biases

  • Actively challenge assumptions
  • Remain receptive to alternative interpretations of the data
  • Look at a belief we hold, and search out ways in which we are wrong
  • This approach is called negative hypothesis testing

 

Diagram showing two hypothesis

Conquering Data Bias

Addressing reporting bias

Hand and scale illustrating ethical conduct

  • Organizations should foster a culture of transparency, accountability, and ethical conduct

Standardized reporting protocols

  • Implementing standardized reporting protocols
  • This can help ensure that data reporting is impartial and comprehensive
Conquering Data Bias

Decision-making processes

  • Implementing structured decision-making processes that encourage critical thinking and diverse perspectives
    • Conducting peer reviews
    • Seeking feedback on analysis methodologies

Two persons doing peer reviews

Conquering Data Bias

Combating bias in algorithms

Person assessing data for representativeness

  • Organizations should prioritize fairness, transparency, and accountability in algorithm development and deployment
  • Rigorously assessing training data for representativeness and diversity
  • Evaluating algorithm performance across diverse demographic groups
  • Promoting ethical awareness among data scientists
Conquering Data Bias

Bias-aware algorithm design

  • Thoughtful feature engineering
    • Selecting and crafting features in an algorithm that consider different perspectives and attributes
  • Fairness constraints
    • Incorporating fairness considerations directly into the model training process
    • For example, an algorithm ensuring a similar distribution across gender, or income categories

Feature engineering and fairness constraints

Conquering Data Bias

Exposing algorithmic bias

Adversarial training

  • Adversarial training
    • Involves training models against adversarial examples specifically designed to expose and mitigate bias

Bias audit

  • Bias audits
    • Involves systematically evaluating models for bias using specialized techniques and metrics (e.g demographic parity)
Conquering Data Bias

Incorporating explainable AI

  • Explainable AI plays an important role in mitigating bias
  • It can provide insights into how algorithms arrive at their decisions
  • Data users can identify and address potential biases more effectively

Robot explaining how it works

Conquering Data Bias

Integrating mitigation strategies

Umbrella for mitigating bias

  • Organizations should adopt a holistic approach that integrates strategies for addressing these biases
  • This is how they can enhance the integrity, reliability, and fairness of their data analyses
Conquering Data Bias

Let's practice!

Conquering Data Bias

Preparing Video For Download...