Structure your answer as a logical pipeline: start with data inspection (types, missing values), then univariate analysis (scales, outliers, statistics), followed by multivariate analysis (correlation heatmap), and conclude with a synthesized summary of data quality issues and actionable next steps. Emphasize reproducibility and clear communication of findings, as Citadel values rigorous, efficient analysis.
Pro tip: Always relate EDA findings to the business problem or modeling goal—don't just list statistics. For example, highlight how outliers or missing values could bias a predictive model and propose specific handling strategies.
Use df.info(), df.dtypes, and df.head() to understand column types, memory usage, and spot obvious issues like numeric columns stored as objects.
Compute missing value counts and percentages per column, and generate descriptive statistics (df.describe()) for numerical and categorical features to understand distributions and scales.
For numerical features, compare ranges and standard deviations; use boxplots or IQR/z-score methods to flag potential outliers, noting their proportion and potential impact.
Generate a correlation heatmap (e.g., sns.heatmap(df.corr())) to identify highly correlated features and potential multicollinearity, and optionally pairplots for key variables.
Write a concise summary of key issues (missingness patterns, outliers, skewness, multicollinearity) and propose concrete next steps such as imputation, transformation, feature selection, or further data collection.
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.