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
Addressing reporting bias

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

- Implementing standardized reporting protocols
- This can help ensure that data reporting is impartial and comprehensive
Decision-making processes
- Implementing structured decision-making processes that encourage critical thinking and diverse perspectives
- Conducting peer reviews
- Seeking feedback on analysis methodologies
Combating bias in algorithms
- 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
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
Exposing algorithmic bias

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

- Bias audits
- Involves systematically evaluating models for bias using specialized techniques and metrics (e.g demographic parity)
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
Integrating mitigation strategies

- 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
Let's practice!
Conquering Data Bias
Preparing Video For Download...