Data Analytics Exploratory Data Analysis and Visualization Syllabus

What is feature engineering?

The process of creating, transforming, or selecting variables from raw data to improve model performance and better expose underlying patterns to algorithms.

Distinguish normalization from standardization in feature scaling.

Normalization (min-max) rescales to a fixed range, usually $[0,1]$: $x' = \frac{x - x_{\min}}{x_{\max} - x_{\min}}$. Standardization rescales to zero mean and unit variance: $x' = \frac{x - \mu}{\sigma}$.

What is one-hot encoding and why is it used?

A technique that converts a categorical variable with $k$ levels into $k$ binary (0/1) indicator columns, so nominal categories can be used by models without implying false ordinal relationships.

Why apply a log transformation to a feature?

To reduce right-skew, compress large ranges, stabilize variance, and linearize multiplicative relationships, making the distribution more symmetric.

What is binning (discretization) in feature engineering?

Converting a continuous variable into discrete categorical intervals (bins), which can capture nonlinearity and reduce the impact of minor fluctuations or outliers.

What is visual encoding in data visualization?

The mapping of data values onto visual properties (position, length, angle, area, color, shape) so that quantitative or categorical information becomes perceivable.

Rank the most to least accurate visual encodings for quantitative data (per Cleveland & McGill).

Position along a common scale is most accurate, followed by position on non-aligned scales, length, angle/slope, area, volume, and color/shading being least accurate.

What is preattentive processing and why does it matter for charts?

The near-instant (under ~250 ms) visual detection of features like color, orientation, or size before conscious attention; using such channels makes key data points 'pop out' effortlessly.

State the data-ink ratio principle (Tufte).

Maximize the proportion of ink used to display actual data relative to total ink; remove non-informative 'chartjunk' to improve clarity.

Which chart type best shows the trend of a continuous variable over time?

A line chart, because the connected line emphasizes continuity and the direction of change across ordered time points.

Which chart type best compares a numeric value across discrete categories?

A bar (or column) chart, since length/position along a common baseline gives accurate category comparisons.

Which chart is appropriate for showing the distribution of a single continuous variable?

A histogram (or density plot); a box plot is used to compare distribution summaries across groups.

Which chart best reveals the relationship between two continuous variables?

A scatter plot, and a bubble chart if a third quantitative variable is encoded by point size.

Why are pie charts often discouraged, and what is their main valid use?

Because humans judge angles/areas poorly, making comparisons hard; they are acceptable only for showing a few parts summing to a meaningful whole (proportions of 100%).

When should a sequential versus a diverging color scale be used?

Sequential scales suit ordered data progressing from low to high; diverging scales suit data with a meaningful midpoint (e.g., zero), emphasizing deviation in two directions.

Why should rainbow (jet) color maps generally be avoided?

They are not perceptually uniform, create false boundaries, are not colorblind-friendly, and misrepresent the ordering of magnitudes.

What is a key guideline for axis (scale) design on bar charts?

The value axis should start at zero, because bar length encodes magnitude; a truncated axis exaggerates differences and misleads the reader.

How can a truncated or nonzero y-axis mislead viewers?

By exaggerating small differences so minor changes appear dramatic; it distorts the proportional comparison the chart implies.

What is 'chartjunk' and why is it harmful?

Non-essential decorative elements (heavy gridlines, 3D effects, textures, clip art) that add no information and distract from or distort the data.

Why can dual y-axes produce misleading charts?

Because the two independent scales can be adjusted arbitrarily to imply a relationship or correlation between series that does not actually exist.

What is the difference between a bar chart and a histogram?

A bar chart compares discrete categories with gaps between bars, while a histogram shows the frequency distribution of a continuous variable with adjacent bars over ordered numeric bins.

When is a stacked bar chart appropriate, and its main drawback?

Appropriate for showing part-to-whole composition within each category; its drawback is that only the bottom segment shares a common baseline, making middle segments hard to compare across bars.

What does an area chart add over a line chart, and when can it mislead?

It fills the region under the line to emphasize volume/magnitude over time; stacked area charts can mislead because upper bands do not share a baseline, distorting individual series.

What is a bubble plot and what dimensions can it encode?

An extension of a scatter plot where a third variable is mapped to marker size (area), encoding three quantitative dimensions (x, y, size), plus optionally color for a fourth.

Why should bubble size encode area rather than radius?

Because human perception judges area, so if radius scales linearly with value the area grows with the square, exaggerating large values: $A = \pi r^{2}$.

What are the five statistics summarized by a box-and-whisker plot?

The minimum, first quartile $Q_1$, median $Q_2$, third quartile $Q_3$, and maximum (whiskers typically extend to $1.5\,\text{IQR}$, with points beyond drawn as outliers).

How does bin width affect a histogram's interpretation?

Too few/wide bins oversmooth and hide structure; too many/narrow bins add noise and spurious spikes. Bin width must balance detail against clarity.

What does a heatmap visualize and what are its axes/values?

A heatmap uses color intensity in a grid to represent the magnitude of a value across two categorical/ordered dimensions (rows and columns), e.g., a correlation matrix or activity by day/hour.

What does a treemap display and how does it encode hierarchy?

A treemap shows hierarchical part-to-whole data using nested rectangles whose area is proportional to a quantitative value, with nesting and color representing categories/levels.

In a BI tool, what does 'connecting to a data source' typically involve, and what is the difference between a live connection and an extract?

Establishing a connection to data (file, database, cloud, API). A live connection queries the source in real time for up-to-date data; an extract imports a static snapshot into the tool for faster, offline performance.

Distinguish dimensions from measures in analytics tools like Tableau.

Dimensions are qualitative/categorical fields (e.g., region, date, product) used to slice and group data; measures are quantitative numeric fields (e.g., sales, profit) that are aggregated. Dimensions define the level of detail; measures are the values summarized.

What is a calculated field versus a parameter in a BI tool?

A calculated field is a new field derived by a formula from existing fields (e.g., profit ratio); a parameter is a single user-adjustable input value that can feed calculations, filters, or reference lines to make views interactive.

What is the difference between a filter and a set in a BI tool?

A filter restricts which rows are displayed based on conditions, whereas a set is a custom, reusable subset of dimension members (defined by condition, top-N, or manual selection) that can be reused across calculations and views to compare 'in' versus 'out' members.