Data Science
This domain tests the end-to-end practice of turning data into defensible decisions. CET questions cover the data lifecycle, data types, descriptive statistics, sampling and bias, visualization choice, and basic inference.
Core rules and concepts
- Lifecycle (CRISP-DM style): business understanding, data understanding, data preparation, modeling, evaluation, deployment. Preparation usually consumes the most effort.
- Data types: nominal (categories, no order), ordinal (ordered categories), interval (equal intervals, no true zero), ratio (true zero). These determine which statistics and charts are valid.
- Descriptive vs inferential: descriptive summarizes the sample; inferential generalizes to a population with uncertainty.
- Center and spread: mean is sensitive to outliers; median is robust. Report spread (standard deviation, IQR) with center.
- Sampling: random sampling reduces selection bias; convenience samples and survivorship bias distort conclusions.
- Inference: a p-value is the probability of data at least this extreme if the null hypothesis is true, not the probability that the null is true.
- Visualization: bar charts compare categories; line charts show trends over time; scatterplots show relationships; histograms show distributions.
Worked example
A churn dataset has 18% missing income values and a 12% churn rate. A model using only complete rows scores well, but the team wants a deployable model.
Reasoning: dropping 18% of rows can remove a non-random group, perhaps high-income or new customers, creating selection bias and weakening deployment. First check whether missingness relates to churn. If it is random, median imputation with a missing-indicator flag preserves rows; if it is informative, model missingness explicitly. Evaluate with precision/recall or AUC, not accuracy, because 88% accuracy is achievable by always predicting no churn. Split before imputing to avoid leakage.
Traps
Do not average ordinal codes as if they were interval data. Do not interpret a p-value as the probability the hypothesis is true. Do not say correlation proves causation. Do not drop missing data without checking the pattern. A colorful chart is not automatically the right chart.