BUSINESS SYSTEMS
│
├── SQL tables
├── CSV/Excel
├── APIs and JSON
├── Logs
├── Documents
└── Images, audio and video
│
▼
DATA COLLECTION
│
▼
DATA VALIDATION
├── Missing values?
├── Duplicates?
├── Invalid ranges?
├── Inconsistent categories?
├── Sensitive data?
└── Future information leaked?
│
▼
DATASET DESIGN
├── Identifier: identifies the row
├── Features: inputs available at prediction time
└── Label: correct historical outcome
│
▼
MODEL DEVELOPMENT
├── Training data
├── Validation data
└── Test data
│
▼
MODEL
│
▼
NEW FEATURES → PREDICTION
Remember
Database table ≠ automatically valid training dataset
Available column ≠ useful feature
High data volume ≠ high data quality
Historical outcome ≠ future feature