Data Analytics: formula sheet

Key formulas

Core idea of data analytics

Raw data → Cleaning → Analysis → Insight → Decision

Use this chain to define the subject and to structure any descriptive answer.

BI vs analytics vs data science (focus)

BI = reporting what happened; Analytics = explaining and forecasting using data; Data science = building predictive models and algorithms

These are differences of focus. The three overlap in real work, so do not call them completely separate.

Data to information

Data + Context and processing = Information; Information + Experience and judgement = Knowledge

Useful for a one-line distinction between data and information in introductory answers.

Three-way classification by format

Data = Structured + Semi-structured + Unstructured

Classify by whether the data follows a fixed schema (structured), carries tags without a rigid schema (semi-structured) or has no model (unstructured).

Core Vs of big data

Volume, Velocity, Variety

These three are the classic characteristics. Always give them first.

Extended Vs of big data

Veracity (quality and trust), Value (usefulness)

Add these for a 5 Vs answer. Mention that some sources list more Vs.

Classification of sources

Internal vs External; Primary vs Secondary

Two separate splits. Internal and external depend on location. Primary and secondary depend on who collected the data first.

Descriptive analytics

Question: What happened?

Summarises historical data using reports, dashboards and statistics. Hindsight.

Diagnostic analytics

Question: Why did it happen?

Finds causes through drill-down, correlation and root-cause analysis. Understanding.

Predictive analytics

Question: What is likely to happen?

Uses statistical and machine learning models. Output is a forecast or probability. Foresight.

Prescriptive analytics

Question: What should we do?

Recommends actions using optimisation, simulation and rules, usually building on predictions. Action.

Lifecycle sequence

Question → Collection → Cleaning → Processing → Analysis → Visualisation → Interpretation → Action and review

Learn the order. Some books merge or rename stages, so state the stage by its function.

Data cleaning tasks

Remove duplicates + treat missing values + correct errors + standardise formats + handle outliers

Use this checklist when asked how to clean data.

Garbage in, garbage out

Quality of output depends on quality of input data

Use it to explain why cleaning cannot be skipped.

Data mining vs data warehousing

Data warehouse = store of integrated historical data; data mining = discovery of patterns from data

Mining often uses a warehouse as its source, but the two are different things.

Supervised vs unsupervised learning

Supervised = labelled data, known outcome; Unsupervised = unlabelled data, finds structure

Classification and regression are supervised. Clustering is unsupervised.

Chart selection rule

Trend over time → line; comparison → bar; share of whole → pie; relationship → scatter

Use as a guide when asked which visual suits a given purpose.

Technique vs tool

Technique = method (clustering); Tool = software (Python, Tableau)

Do not list a tool when the question asks for a technique.

Four types of analytics

Descriptive → Diagnostic → Predictive → Prescriptive

Use this ladder to classify any application: what happened, why, what next, what to do.

Application answer structure

Area + Data used + Technique + Governance benefit

A simple pattern for every answer on applications.

Exception test

Exception = record that breaks a defined rule or normal pattern

Basis of fraud and compliance analytics, for example duplicate invoice numbers or payments just below an approval limit.

Key roles under the DPDP Act

Data Principal = the individual; Data Fiduciary = decides purpose and means; Data Processor = processes on behalf of the Fiduciary

Always identify the role of each party first in a case question.

Grounds for processing

Lawful purpose = consent OR legitimate use permitted by the Act

Consent must be free, specific, informed, unconditional and unambiguous.

Core duties of a Data Fiduciary

Notice + purpose limit + accuracy + security safeguards + breach reporting + erasure when purpose ends

Use this list as a checklist for any analytics scenario.

Rights of a Data Principal

Access + correction and erasure + grievance redressal + nomination + withdrawal of consent

Withdrawal of consent is easy to forget, so add it to every list of rights.

Ethical principles in analytics

Fairness + transparency + accountability + security + privacy

Use these as headings when the question asks about ethics.

Quick revision

• Data analytics means examining data to find patterns and support decisions.
• Data can be structured, semi-structured or unstructured; know an example of each.
• Know primary versus secondary sources and internal versus external sources.
• Descriptive analytics answers what happened.
• Diagnostic analytics answers why it happened.
• Predictive analytics estimates what is likely to happen.
• Prescriptive analytics recommends what to do.
• The lifecycle runs from defining the problem through collecting, cleaning, analysing and presenting to acting on results.
• Poor data quality leads to poor conclusions, so cleaning is a core step.
• Match each tool or technique to a lifecycle stage.
• Corporate uses include risk, compliance, audit, fraud detection and reporting.
• Legal and ethical issues include privacy, consent, security, bias and transparency.

Common mistakes

• Treating data analytics, data science and BI as the same thing. Fix: Remember the focus of each: BI reports the past, analytics explains and forecasts, data science builds predictive models. Give at least three comparison points.
• Writing only a definition and no importance or examples. Fix: Add business and governance uses with a concrete example, such as detecting unusual transactions or tracking compliance.
• Calling emails, JSON and XML all unstructured, or all semi-structured. Fix: Emails as free text, images and video are unstructured. JSON, XML and HTML carry tags, so they are semi-structured.
• Listing only three Vs when the question says 5 Vs, or the other way round. Fix: Follow the number in the question. For five, give volume, velocity, variety, veracity and value. Name the extra ones if asked for more.
• Calling a sales report that shows last quarter's figures predictive analytics. Fix: Ask whether it looks backward or forward. A summary of past data is descriptive.
• Treating diagnostic and descriptive analytics as the same. Fix: Descriptive states what happened. Diagnostic explains why, by drilling into causes.
• Jumping straight to analysis and skipping cleaning. Fix: State that cleaning and preparation come first and that poor data gives unreliable results.
• Mixing up cleaning and processing. Fix: Cleaning fixes errors and gaps. Processing changes the form or structure, such as merging, converting or aggregating.
• Treating data mining and data warehousing as the same thing Fix: Remember: warehouse stores integrated historical data; mining analyses data to find patterns.
• Listing tools like Python or Tableau when asked for techniques Fix: Techniques are methods such as classification and clustering. Tools are software.

Exam tips

• For 'distinguish' questions, write a point-wise comparison with at least four points and add a one-line note that the terms overlap.
• Always add one governance or compliance use, since this is a paper for company secretaries.
• Use the chain data to insight to decision as a ready-made definition structure.
• Close with a risk line on data quality, privacy or security to show practical awareness.
• Keep the answer short and structured with headings or bullet points, as time is limited.
• For 'distinguish' questions, use a short two-column comparison with at least four points. It scores better than long paragraphs.
• Give the Vs in a fixed order and say how many you are listing. If the question does not give a number, give five and note that some texts add more.
• Attach an Indian corporate example to every category. It shows application, not rote learning.