Machine learning is easier to navigate by the shape of the problem than by the name of the method. Each playbook below takes one problem class through the same ten steps, from the theory it leans on to what to watch once it ships, and links every step to the topics and lectures it needs. Pick the one closest to what you are building.
Structured data
- Binary classificationchurn, fraud, loan default, lead conversion
- Regressionprice, lifetime value, delivery time, energy load
- Survival analysistime to churn, machine failure, loan default, patient outcome
- Time series forecastingdemand, traffic, revenue, capacity
- Clusteringcustomer segments, product groupings, cohort discovery
- Anomaly detectionfraud alerts, system monitoring, quality control
- Entity resolutiondeduplicate customers, match products across suppliers, link records
Decide & act
Recommendation & search
Deep learning
- Text classificationticket routing, sentiment, content moderation, intent
- Information extractioninvoices, contracts, resumes, forms
- Image classificationproduct photos, defect detection, medical imaging, moderation
- Speech & audio recognitioncall transcription, keyword spotting, sound events
- Image generationproduct imagery, data augmentation, style transfer
- Language model trainingpretrain a small model, adapt an open model to a domain, preference tuning