NCA-ADS - Descriptive Analysis and Visualization (13% of the exam) - Section 4.1
Perform exploratory data analysis and descriptive statistics.
Perform EDA by computing summary statistics - mean, median, standard deviation, quantiles, and correlation coefficients - to understand data distributions and relationships before modelling. Use these findings to inform preprocessing decisions such as scaling, encoding, and outlier treatment.
EDA
Practice question for this objective
lock_openFree sampleDescriptive Analysis and Visualizationeasy
A data scientist has just loaded a 2-million-row cuDF DataFrame of website session logs and wants a fast first look at the data. As the opening step of exploratory data analysis, what is the primary purpose of generating per-column summary statistics such as count, mean, minimum, maximum, and the quartiles?
ATo understand the central tendency, spread, and range of each numeric column and to surface anomalies such as impossible values or missing entries before modelling.check_circle Correct
BTo select the final set of predictive features by ranking columns according to their correlation with the target variable.
CTo train a baseline regression model whose coefficients quantify each feature's contribution to the outcome.
DTo permanently impute every missing value using the column mean so the dataset is immediately ready for production.
Descriptive summary statistics are an early EDA step that characterises each column's distribution and exposes data-quality problems before modelling. Per-column descriptive statistics report central tendency and spread and expose out-of-range or missing values, giving an analyst a quick, model-free understanding of the data and its quality before any feature engineering or training occurs.
Why A is correct: Correct. Summary statistics are an early EDA step whose purpose is to characterise distributions and reveal data-quality issues, for example a negative session duration or a count below the row total signalling nulls, before any modelling begins.
Why B is wrong: Tempting because feature selection is also part of a data pipeline, but summary statistics describe individual columns and do not measure relationships to a target, so they cannot rank predictive power on their own.
Why C is wrong: Tempting because a baseline model is a reasonable later step, but computing descriptive statistics fits no model and produces no coefficients, so it cannot establish a baseline predictor.
Why D is wrong: Tempting because the mean is part of the summary, but reporting statistics does not modify the data, and blanket mean imputation is a deliberate later choice rather than the purpose of describing the data.
Exam traps in Descriptive Analysis and Visualization
Answers that look right on this material and are not. Each one is a distractor from a different question in the NCA-ADS bank for this domain.
Feature A's values are, on average, more tightly clustered around their mean than feature B's values.
errorWhy it is wrong: Tempting if the relationship is remembered in reverse, but a larger standard deviation means more dispersion, not tighter clustering, so this describes the opposite of feature A.
The value_counts operation, which returns frequency counts for each unique entry in every column
errorWhy it is wrong: value_counts tallies occurrence frequencies for categorical-style data and does not compute numeric summary statistics such as mean, standard deviation, or quartiles.
Calling the info method on the DataFrame, which returns count, mean, standard deviation, and the quartiles for each numeric column.
errorWhy it is wrong: The info method is tempting because it is an early EDA call, but it reports column names, non-null counts, and dtypes only. It does not compute the mean, standard deviation, or quartiles.