CertNexus CDSP Exam Syllabus

Defining the need to be addressed through the application of data science (7-9%)

• Milestone
• POC (Proof of concept)
• MVP (Minimal Viable Product)

- Become aware of data privacy, security, and governance policies

• GDPR
• HIPPA
• California Privacy Act

- Obtain permission/access to stakeholder data
- Ensure appropriate voluntary disclosure and informed consent controls in place

• Optimization problem
• Forecasting problem
• Regression problem
• Classification problem
• Segmentation/Clustering problem

- Identify data sources and type

• Structured/unstructured
• Image
• Text
• Numerical
• Categorical

- Select modeling type

• Regression
• Classification
• Forecasting
• Clustering
• Optimization
• Recommender systems

• Write a query for a SQL database
• Write a query for a NoSQL database
• Read data from/write data to cloud storage solutions
  1. AWS S3
  2. Google Storage Buckets
  3. Azure Data Lake

- Become aware of first-, second-, and third-party data sources

• Understand data collection methods
• Understand data sharing agreements, where applicable

- Explore third-party data availability

• Demographic data
• Bloomberg

- Collect open-source data

• Use APIs to collect data
• Scrape the web

- Generate data assets

• Dummy or test data
• Randomized data
• Anonymized data
• AI-generated synthetic data

• Nulls
• Duplicates
• Corrupt values

- Parse the data
- Check for corrupted data
- Correct the data format
- Deduplicate data
- Apply risk and bias mitigation techniques

• Understand common forms of ML bias
  1. Sampling bias
  2. Measurement bias
  3. Exclusion bias
  4. Observer bias
  5. Prejudicial bias
  6. Confirmation bias
  7. Bandwagoning

- Identify the sources of bias

• Sources of bias include data collection, data labeling, data transformation, data imputation, data selection, and data training methods
• Use exploratory data analysis to visualize and summarize the data, and detect outliers and anomalies
• Assess data quality by measuring and evaluating the completeness, correctness, consistency, and currency of data
• Use data auditing techniques to track and document the provenance, ownership, and usage of data, and applied data cleaning steps

- Mitigate the impact of bias

• Apply mitigation strategies such as data augmentation, sampling, normalization, encoding, validation

- Evaluate the outcomes of bias

• Use methods such as confusion matrix, ROC curve, AUC score, and fairness metrics

- Monitor and improve the data cleaning process

• Establish or adhere to data governance rules, standards, and policies for data and the data cleaning process

• Make sure a common key exists in all datasets
• Unique identifiers

- Load data

• Load into DB
• Load into dataframe
• Export the cleaned dataset
• Load into visualization tool

- Make an endpoint or API

• Word2vec
• TF-IDF
• Glove

- Generate latent representations for image data

• One-hot encoding
• Target encoding
• Label encoding or Ordinal encoding
• Dummy encoding
• Effect encoding
• Binary encoding
• Base-N encoding
• Hash encoding

- Split features

• Text manipulation
  1. Split
  2. Trim
  3. Reverse
• Manipulate data
• Split names
• Extract year from title

- Convert dates to useful features
- Apply feature reduction methods

• PCA
• t-SNE
• Random forest
• Backward feature elimination
• Forward feature selection
• Factor analysis
• Missing value ratio
• Low-variance filter
• High-correlation filter
• SVD
• False discovery rate
• Feature importance methods

• Regression
  1. Linear regression
  2. Random forest
  3. XGBoost

- Classification

• Logistic regression
• Random forest classification
• XGBoost classifier
• naïve Bayes

- Forecasting

• ARIMA

- Clustering

• k-means
• Density-based methods
• Hierarchical clustering

- Train model or pre-train or adapt transformers
- Tune hyper-parameters, if applicable

• Cross-validation
• Grid search
• Gradient decent
• Bayesian optimization

• Confusion matrix
• Learning curve

- Select best-performing model
- Store model for operational use

• MLflow
• Kubeflow

• Experimental design
  1. Design use cases
  2. Test creation
  3. Statistics

- Define success criteria for test
- Evaluate test results

• AWS SageMaker
• Azure ML
• Docker
• Kubernetes

- Ensure model works operationally
- Monitor pipeline for performance of model over time

• MLflow
• Kubeflow
• Datadog

- Consider enterprise data strategy and data management architecture to facilitate the end-to-end integration of data pipelines and environments

• Data warehouse and ETL process
• Data lake and ETL processes
• Data mesh, micro-services, and APIs
• Data fabric, data virtualization, and low-code automation platforms

• Web frameworks (Flask, Django)
• Basic HTML
• CSS

- Derive insights from findings
- Identify features that drive outcomes (e.g., explainability, interpretability, variable importance plot)
- Show model results
- Generate lift or gain chart
- Ensure transparency and explainability of model

• Use explainable methods (e.g., intrinsic and post hoc)
  1. Visualization
  2. Feature importance analysis
  3. Attention mechanisms
  4. Avoiding black-box techniques in model design
  5. Explainable AI (XAI) frameworks and tools
  - SHAP
  - LIME
  - ELI5
  - What-If Tool
  - AIX360
  - Skater
  - Et al

- Document the model lifecycle

• ML design and workflow
• Code comments
• Data dictionary
• Model cards
• Impact assessments

- Engage with diverse perspectives

• Stakeholder analysis
• User testing
• Feedback loops

- Participatory design

• Implement self-service data/analytics platforms

- Create a culture of data literacy

• Educate employees on how to use data effectively
• Offer support and guidance on data-related issues
• Promote transparency and collaboration around data