The part that tripped me up wasn't the modeling, it was remembering to actually narrate.
Start by clarifying the problem and success metrics, then perform quick EDA to understand data quality and distributions. Build a simple baseline model first, iterate with more complex models, and evaluate using appropriate metrics while explaining trade-offs. Throughout, narrate your reasoning and link decisions to business impact.
Pro tip: Verbalize your thought process and trade-offs (e.g., why you choose a metric or model) rather than silently coding; interviewers assess your reasoning as much as your results.
Ask about the prediction target, business objective, and evaluation criteria. Quickly inspect the dataset shape, types, missing values, and target distribution.
Analyze feature distributions, correlations, and potential leakage. Visualize key relationships and note data quality issues to address.
Handle missing values, encode categorical variables, scale numerical features, and create relevant features. Split data into train/validation/test sets.
Start with a simple baseline (e.g., logistic regression). Then try more complex models (e.g., tree-based ensembles), tune hyperparameters, and compare performance.
Evaluate using appropriate metrics (e.g., AUC, F1) and discuss trade-offs. Summarize findings, limitations, and potential next steps.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.