Getting the right data

Artificial Intelligence (AI) Strategy

Vidhi Chugh

AI strategist and ethicist

Data availability

• More data -> better model outcomes

• Not just the data quantity
• Labeling, quality, and timeliness

• Model captures underlying patterns

• Data-centric science
• Systematically engineering the data

Data relevance

 

• Richness and relevance of patterns
• Irrelevant data misleads the model -> lowers accuracy

 

• Example: credit scoring model
  • Relevant attributes
  • Transactions history
  • Assets profile

 

Time relevancy

• Supply chain dynamics have changed post pandemic

Data privacy

 

• Sensitive user data

• Data privacy standards, such as, GDPR

 

• Ethical practices

• Enables user trust

Data dictionary

 

• Gives the meaning of different data fields and their significance

 

• Domain experts link data with business decisions

 

• Redundant data does not add value to model predictions

 

Data sampling

• When working with more data
• Create a sampled dataset
• Comparable results while being economical

Data augmentation

• But, what if data is not sufficient?
• Augment data - create new records from existing datasets

Data diversity

 

• Create reliable models

• Good model accuracy

• Example: Loan application

  • Include different age groups and ethnicities

 

Data quality

 

• Complete and comprehensive
• Accurate data; age and date of birth

 

• Missing data
• Correct data labels

Let's practice!

Artificial Intelligence (AI) Strategy