COGNIZANT DATA ANALYST

INTERVIEW QUESTIONS
0-3 YOE

1) How do you approach cleaning and preparing data for

analysis?
Answer:

Cleaning and preparing data is a critical first step in the data analysis process, ensuring
data is accurate, consistent, and usable. Here’s how I typically approach it:

Step-by-Step Approach:

1. Understand the Dataset:

o Review column names, data types, and the number of unique values.

o Use methods like .info(), .describe(), .head() in Python or schema review in

SQL.

2. Remove Duplicates:

o Check for and remove duplicate records using drop_duplicates() in pandas or

DISTINCT in SQL.

3. Handle Missing Values:

o Identify missing values using .isnull().sum() or COUNT(*) WHERE column IS

NULL.

o Handle them by:

▪ Dropping rows/columns (if they have excessive missing values).

▪ Imputing with mean, median, or mode.

▪ Forward/backward fill for time-series data.

 4. Correct Data Types:

o Convert columns to appropriate types (e.g., dates to datetime, numeric

strings to integers).

5. Standardize & Normalize Data:

o Ensure consistent formatting (e.g., ‘Yes’/‘No’ vs. 1/0).

o Apply normalization or scaling techniques if required for ML models.

6. Fix Inconsistencies:

o Fix inconsistent category labels like “Male”, “M”, “male” → “Male”.

7. Feature Engineering (Optional):

o Create new columns, such as extracting month from a date or computing

total_price = quantity * price.

8. Outlier Detection:

o Use statistical methods (e.g., IQR, Z-score) or visualizations (boxplots) to

detect and handle outliers.

Tools & Libraries Used:

• Python (pandas, numpy, scipy)

• SQL (Cleansing at database level)

• Excel (for initial review if needed)

2) Explain how SQL is used in data analysis.

Answer:

SQL (Structured Query Language) is a fundamental tool in data analysis, especially when
working with relational databases.

Use of SQL in Data Analysis:

1. Data Extraction:

o SQL helps retrieve data from databases using SELECT statements.

2. Data Filtering and Sorting:

 o Use WHERE, AND, OR, LIKE, BETWEEN, IN to filter data.

o Use ORDER BY to sort data for better visibility and understanding.

3. Aggregations & Grouping:

o Perform summary statistics using COUNT(), SUM(), AVG(), MAX(), MIN().

o Combine with GROUP BY for category-wise analysis.

4. Joining Tables:

o Use INNER JOIN, LEFT JOIN, etc., to merge data from multiple tables.

o Essential when working with normalized data spread across tables.

5. Window Functions:

o Use ROW_NUMBER(), RANK(), LEAD(), LAG() for advanced calculations

without collapsing rows.

6. Data Cleaning in SQL:

o Handle null values with IS NULL, COALESCE().

o Replace strings using REPLACE() or CASE WHEN statements.

7. Subqueries and CTEs:

o Use nested queries and WITH statements for better readability and
reusability.

8. Data Validation and Checks:

o Check duplicates, data ranges, invalid entries directly in SQL before analysis.

Real-Life Scenario:

Suppose you're working with a sales database — SQL helps in calculating:

SELECT region, SUM(sales) AS total_sales

FROM orders

WHERE order_date BETWEEN '2024-01-01' AND '2024-03-31'

GROUP BY region

ORDER BY total_sales DESC;

This gives regional sales trends, a typical analyst task.

3) How do you handle missing or incomplete data in a dataset?

Answer:

Handling missing or incomplete data is crucial to ensure the accuracy and robustness of
your analysis.

Approach to Handling Missing Data:

1. Identify Missing Values:

o In Python: df.isnull().sum()

o In SQL: SELECT COUNT(*) FROM table WHERE column IS NULL

2. Understand the Pattern:

o Determine if the data is missing at random (MAR), missing completely at

random (MCAR), or not at random (MNAR).

3. Common Techniques to Handle Missing Data:

Technique Use Case

Drop rows/columns When missing values are very few and insignificant

Mean/Median Imputation For numeric data, when distribution is symmetric/skewed

Mode Imputation For categorical variables

Forward/Backward Fill For time-series or sequential data

Predictive Imputation Use ML models to predict missing values

Constant Value Imputation Assign values like “Unknown” or 0

4. Flag Missing Data (Optional):

o Create a new boolean column to mark missing entries, useful in ML models.

5. Document Assumptions:

o Keep track of what and how you imputed or removed data for reproducibility.
 Example in Python:

# Impute numeric column with mean

df['Age'].fillna(df['Age'].mean(), inplace=True)

# Fill categorical column with mode

df['Gender'].fillna(df['Gender'].mode()[0], inplace=True)

Important Consideration:

Always assess the impact of missing data on your analysis before deciding on a strategy.
Sometimes, missingness itself can be an insight (e.g., customers not responding to a
question).

4) What is data normalization, and why is it important?

Answer:

What is Data Normalization?

Data normalization refers to the process of transforming data into a common scale without
distorting differences in the ranges of values. It ensures that each feature contributes
equally to the analysis or model.

It can also refer to a database design technique that organizes tables to reduce
redundancy and improve data integrity.

Types of Normalization (in Data Preprocessing):

1. Min-Max Normalization:

o Scales values between 0 and 1.

o Formula:

x′=x−min(x)/(max(x)−min(x))

2. Z-score Normalization (Standardization):

 o Scales data based on mean = 0 and standard deviation = 1.

o Formula:

x′=(x−μ)/σ

3. Decimal Scaling:

o Moves the decimal point of values to bring them within a standard range.

Why is Normalization Important?

• Removes Bias Due to Scale:

o Algorithms like KNN, SVM, and Gradient Descent are distance-based and
sensitive to scale.

• Improves Model Accuracy:

o Balanced features lead to faster convergence and better results in machine

learning.

• Enhances Data Integrity (in Database Normalization):

o Prevents anomalies, ensures data is stored efficiently, and maintains

relationships logically.

5) What tools and techniques do you use for data visualization?

Answer:

As a Data Analyst, data visualization is key to presenting findings in a clear, compelling way.

Tools I Use for Visualization:

Tool Use Case

Business dashboards, KPI reports, slicers, DAX-

Power BI
based visuals
Tool Use Case

Interactive dashboards, storyboards, drill-down

Tableau
charts

Excel Quick analysis with pivot charts, bar/line/pie charts

Python (Matplotlib, Seaborn,

Custom data visualizations and EDA plots
Plotly)

SQL (basic charting in some

Data summaries with export to Excel or BI tools
platforms)

Common Visualization Techniques:

Chart Type Purpose

Bar/Column Chart Comparing categories

Line Chart Showing trends over time

Pie/Donut Chart Showing proportions (use sparingly)

Histogram Understanding distribution of numerical data

Boxplot Identifying outliers and data spread

Heatmap Visualizing correlation or matrix-based data

Scatter Plot Finding relationships between two variables

TreeMap/Sunburst Hierarchical data representation

Best Practices:

• Choose the right chart for the right data.

• Avoid cluttered visuals.

 • Use consistent color schemes.

• Highlight key insights with annotations.

6) Explain the difference between correlation and causation.

Answer:

Correlation:

• Definition: A statistical relationship or association between two variables.

• Direction: Can be positive, negative, or zero.

• Measurement: Pearson’s correlation coefficient (r), ranging from -1 to 1.

• Example: Ice cream sales and temperature tend to rise together (positive
correlation).

Causation (Cause and Effect):

• Definition: One variable directly affects the other.

• Requires Evidence: Correlation alone doesn’t prove causation.

• Example: Smoking causes an increase in the risk of lung cancer.

Key Differences:

Feature Correlation Causation

Meaning Variables move together One variable influences the other

Directionality Not implied Implied (cause → effect)

Proof Needed Statistical relationship Statistical + Experimental evidence

Example Sleep hours ↔ Productivity Exposure to virus → Illness

 Important Note:

"Correlation does not imply causation."

Just because two variables move together doesn't mean one causes the other.
Confounding variables or coincidence could be at play.

7) How do you perform hypothesis testing in data analysis?

Answer:

Hypothesis Testing is a statistical method used to make inferences or draw

conclusions about a population based on sample data.

Step-by-Step Approach:

1. Formulate Hypotheses:

o Null Hypothesis (H₀): Assumes no effect or difference.

o Alternative Hypothesis (H₁): Assumes there is an effect or difference.

Example: H₀: New marketing strategy has no effect on sales.

H₁: New marketing strategy increases sales.

2. Choose the Right Test:

o Based on data type and distribution:

▪ Z-test/T-test: Comparing means.

▪ Chi-square test: Categorical variables.

▪ ANOVA: Comparing means of 3+ groups.

▪ Proportion test: For yes/no outcomes.

3. Set the Significance Level (α):

o Common values: 0.05, 0.01.

o Represents the risk of rejecting a true null hypothesis (Type I error).

4. Calculate the Test Statistic and P-value:

o Use statistical formulas or tools (Python, Excel, R).

 5. Decision Making:

o If p-value ≤ α, reject H₀ → evidence supports H₁.

o If p-value > α, fail to reject H₀.

6. Interpret Results in Business Context:

o Clearly communicate findings to stakeholders.

Example in Python (T-test):

from scipy.stats import ttest_ind

# Sales before and after campaign

before = [100, 120, 110, 115]

after = [130, 145, 140, 150]

t_stat, p_val = ttest_ind(before, after)

if p_val < 0.05:

print("Reject H₀ → Campaign has a significant effect")

else:

print("Fail to reject H₀ → No significant difference")

8) What is the difference between a primary key and a foreign key

in databases?
Answer:

Both Primary Key and Foreign Key are used to establish and enforce relationships
between tables in a relational database.
 Feature Primary Key Foreign Key

Uniquely identifies each record in

Purpose Links one table to another
a table

Uniqueness Must be unique and non-null Can have duplicates and nulls

Location Defined within the current table Refers to a primary key in another table

EmployeeID in Salaries referencing

Example EmployeeID in Employees table
Employees

Real-Life Example:

-- Primary Key

CREATE TABLE Employees (

EmployeeID INT PRIMARY KEY,

Name VARCHAR(50)

);

-- Foreign Key

CREATE TABLE Salaries (

SalaryID INT PRIMARY KEY,

EmployeeID INT,

FOREIGN KEY (EmployeeID) REFERENCES Employees(EmployeeID)

);

• EmployeeID is the primary key in Employees.

• The same EmployeeID acts as a foreign key in Salaries to maintain referential

integrity.
9) How do you ensure data accuracy in your analysis?
Answer:

Ensuring data accuracy is essential for making reliable business decisions.

Steps I Follow:

1. Understand Data Source:

o Check data origin, collection method, and frequency.

2. Validate Data Types & Formats:

o Ensure each column has the correct data type (e.g., date columns are in
datetime format).

3. Remove Duplicates and Inconsistencies:

o Use drop_duplicates() or SQL DISTINCT.

o Standardize category values (e.g., “Male”, “male”, “M” → “Male”).

4. Handle Missing/Null Values:

o Identify missing data using .isnull() or IS NULL and impute appropriately.

5. Cross-Check with Source Systems or Business Logic:

o Validate totals, summaries, or group-level data against raw system exports or

reports.

6. Outlier Detection:

o Use boxplots, IQR, or Z-scores to catch abnormal values.

7. Use Checksums & Audit Logs (for large systems):

o Match row counts or hashes when importing/exporting data.

8. Automated Testing:

o Use scripts or data validation tools to enforce rules (e.g., no negative sales
values).

9. Peer Reviews & Feedback:

o Discuss logic and findings with domain experts for verification.

 Tools Used:

• Python (pandas, NumPy, Pytest)

• SQL queries for integrity checks

• Power BI’s data profiling feature

• Excel for quick cross-validation

10 )
𝗪𝗿𝗶𝘁𝗲 𝗮 𝗾𝘂𝗲𝗿𝘆 𝘁𝗼 𝗳𝗶𝗻𝗱 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀 𝘄𝗵𝗼 𝗵𝗮𝘃𝗲 𝗺𝗮𝗱𝗲 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗼𝗻𝗲
𝗽𝘂𝗿𝗰𝗵𝗮𝘀𝗲 𝗼𝗻 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗱𝗮𝘁𝗲𝘀 𝗮𝗻𝗱 𝗰𝗮𝗹𝗰𝘂𝗹𝗮𝘁𝗲 𝘁𝗵𝗲𝗶𝗿 𝘁𝗼𝘁𝗮𝗹 𝗽𝘂𝗿𝗰𝗵𝗮𝘀𝗲
𝗮𝗺𝗼𝘂𝗻𝘁.

Input Table: Sales

sale_id product_id customer_id sale_date sale_amount

1 101 1001 2023-09-01 500

2 102 1002 2023-09-02 300

3 101 1001 2023-09-03 400

4 103 1003 2023-09-04 700

5 102 1002 2023-09-05 600

SELECT

customer_id,

COUNT(DISTINCT sale_date) AS distinct_purchase_days,

SUM(sale_amount) AS total_purchase_amount
FROM

Sales

GROUP BY

customer_id

HAVING

COUNT(DISTINCT sale_date) > 1;

Explanation:

• COUNT(DISTINCT sale_date) → ensures we only count unique dates for purchases.

• HAVING COUNT(DISTINCT sale_date) > 1 → filters customers who purchased on

multiple dates.

• SUM(sale_amount) → gives the total amount spent by each qualifying customer.

Result for your sample data:

customer_id distinct_purchase_days total_purchase_amount

1001 2 900

1002 2 900

Both customers 1001 and 1002 purchased on two different dates.

11) There are two table Table A and Table B.

Table A
7
7
7
7

Table B
7
7
7

How many rows should be as output from below table queries?

A INNER JOIN B =
A LEFT JOIN B =
A UNION B =
A UNION ALL B =
A EXCEPT B =
A INTERSECT B =
A CROSS JOIN B WHERE A.ID=7 =

Table A:

7 → 4 rows (value 7)

Table B:

7 → 3 rows (value 7)
We're assuming there's an implicit ID column in each table with values = 7. Now let's
analyze each SQL operation:

A INNER JOIN B

Each row from A joins with each matching row in B (on value = 7).

• A has 4 rows with 7

• B has 3 rows with 7

Result = 4 × 3 = 12 rows

A LEFT JOIN B

All 4 rows from A will join with matching rows in B (3 rows).

• For each of the 4 rows in A, there are 3 matches in B

Result = 4 × 3 = 12 rows

A UNION B

Removes duplicates (DISTINCT).

All values = 7
Only one unique value = 7

Result = 1 row

A UNION ALL B

Keeps all rows, including duplicates

• 4 from A

• 3 from B

Result = 4 + 3 = 7 rows
 A EXCEPT B

Returns values in A that are not in B

All values in A are 7, and all values in B are 7
Result = 0 rows

A INTERSECT B

Returns common values (distinct) in A and B

Only value = 7 (in both)

Result = 1 row

A CROSS JOIN B WHERE A.ID = 7

Assuming CROSS JOIN is filtered with WHERE A.ID = 7 (and all values in A are 7), it's the
same as:

SELECT *

FROM A

CROSS JOIN B

WHERE A.ID = 7;

All rows in A meet the condition (4 rows), and all 3 in B combine with each → 4 × 3 = 12 rows

Result = 12 rows

Final Answers Recap:

Query Output Rows

A INNER JOIN B 12

A LEFT JOIN B 12

A UNION B 1
Query Output Rows

A UNION ALL B 7

A EXCEPT B 0

A INTERSECT B 1

A CROSS JOIN B WHERE A.ID = 7 12