Predictive analytics and machine learning forecast outcomes from historical patterns. The book covers the supervised / unsupervised split, ML pipelines, MLOps, and Philippine industry use cases.
9.1 Predictive Analytics and Machine Learning
Three learning paradigms
Paradigm
Data needed
Example task
Supervised
Labelled
Classification + regression
Unsupervised
No labels
Clustering + reduction
Reinforcement
Reward signal
Agent decisions
The ML pipeline
Common algorithms by task
Task
Algorithms
Classification
Logistic, random forest, XGBoost, neural net
Regression
Linear, ridge, gradient boosting, neural net
Clustering
k-means, DBSCAN, hierarchical
Recommendation
Matrix factorisation, two-tower neural
Forecasting
ARIMA, Prophet, recurrent / temporal nets
MLOps essentials
Component
Tool examples
Versioning
DVC, MLflow
Pipelines
Vertex AI Pipelines, Kubeflow
Feature store
Feast, SageMaker Feature Store
Model registry
MLflow, SageMaker
Drift monitoring
Evidently AI, Arize, WhyLabs
Risks the EA must address
Risk
Mitigation
Bias
Fairness metrics per protected attribute
Drift
Scheduled retrain + monitoring
Explainability
SHAP / LIME stored per decision
Data leakage
Train/test split discipline
Lineage gaps
DVC + metadata catalogue
Mentor’s tip: Pipeline first, model second. MLOps gives you reproducibility, drift detection, and audit. Fair-lending and explainability are statutory in banking; bake them into the pipeline before deploying.
Discussion
Loading…