Introduction to Data Science for Business Decision-making: formula sheet

Key formulas

Three components of data science

Data science = Statistics and Mathematics + Computing + Domain knowledge

A memory frame for definition answers. Missing any one component weakens the result.

Characteristics of big data (the 3 Vs)

Volume + Velocity + Variety

Some books add Veracity and Value. Name the version your study material uses and list all the Vs you give.

Flow from data to decision

Data → Information → Insight → Decision

Useful for explaining importance and scope in one line.

Life cycle sequence

Business understanding → Data collection/understanding → Data preparation → Exploratory analysis → Modelling → Evaluation → Deployment → Monitoring

Iterative: evaluation or monitoring can send you back to earlier stages.

CRISP-DM phases

Business understanding → Data understanding → Data preparation → Modelling → Evaluation → Deployment

Six phases. Use the grouping your study material gives if it differs slightly.

Three types by format

Structured (fixed schema) | Semi-structured (tags, flexible) | Unstructured (no model)

Give one business example for each. Semi-structured sits between the other two.

Big data 5 Vs

Volume, Velocity, Variety, Veracity, Value

Volume = size, Velocity = speed, Variety = formats, Veracity = quality and trust, Value = usefulness.

Source classification

Internal (inside the firm) | External (outside the firm)

Internal: ERP, sales, payroll. External: government data, market reports, social media.

Primary vs secondary

Primary = first-hand, new collection | Secondary = already collected by others

Based on who collected the data and for what purpose, not on location.

Mean

Mean = Σx ÷ n

Used to fill missing numeric values when data has no extreme outliers.

Median

Middle value of data arranged in order; for even n, average of the two middle values

Preferred for imputation when data is skewed or has outliers.

Mode

Most frequent value

Used to fill missing categorical values such as city or product type.

Interquartile range (IQR)

IQR = Q3 − Q1

Measures the spread of the middle 50% of the data.

IQR rule for outliers

Lower fence = Q1 − 1.5 × IQR; Upper fence = Q3 + 1.5 × IQR

A common rule of thumb: values outside the fences are flagged as possible outliers. Flagged does not mean wrong.

Z-score

z = (x − mean) ÷ standard deviation

Measures how many standard deviations a value is from the mean. An absolute value above 3 is often flagged as a possible outlier.

Min-max scaling

x' = (x − min) ÷ (max − min)

Rescales values to the range 0 to 1.

Missing percentage

Missing % = (Number of missing values ÷ Total records) × 100

Helps decide between deletion and imputation.

Descriptive analytics

Question: What happened? Tools: mean, median, totals, ratios, charts, dashboards

Looks backward. No forecast or recommendation.

Predictive analytics

Question: What is likely to happen? Tools: regression, time series, classification

Uses historical data. Output is a probability or estimate, not a guarantee.

Prescriptive analytics

Question: What should we do? Tools: optimisation, simulation, decision rules

Needs an objective and constraints. Builds on predictions.

Simple linear regression

Y = a + bX

Predicts a numeric value Y from X. Regression shows association; it does not by itself prove cause.

Supervised vs unsupervised learning

Supervised: labelled outcomes. Unsupervised: no labels

Regression and classification are supervised. Clustering is unsupervised.

Quick revision

• Data science combines statistics, computing and domain knowledge to get insight from data for decisions.
• Life cycle runs from problem definition to data collection, preparation, exploration, modelling, interpretation and action.
• Always begin with a clear business problem, since it decides what data you need.
• Structured data fits rows and columns; unstructured data such as text, images and audio does not.
• Primary data is collected first-hand; secondary data already exists from another source.
• Data preparation covers cleaning, handling missing values, removing duplicates and treating outliers.
• EDA summarises and visualises data to spot patterns, trends and anomalies before modelling.
• Descriptive analytics asks what happened; predictive asks what is likely to happen; prescriptive asks what should we do.
• Prescriptive analytics builds on prediction and recommends actions.
• Typical uses include customer segmentation, demand forecasting, fraud detection and pricing.
• Ethical issues include privacy, consent, bias, transparency and data security.
• Good decisions need good data: poor quality input leads to poor output.

Common mistakes

• Treating data science and statistics as the same thing Fix: Say statistics is a theoretical foundation, while data science adds computing, large and unstructured data, and domain knowledge.
• Describing evolution as a list of random terms Fix: Present it as stages in time order, with the driver of each stage such as computers, internet data or cheaper computing.
• Starting the life cycle with modelling or data collection. Fix: Always begin with business understanding: the problem and the decision to be supported.
• Treating the cycle as a one-way straight line. Fix: State that it is iterative and give an example of looping back, such as poor evaluation leading to more data preparation.
• Calling emails unstructured without nuance, or calling all text unstructured. Fix: Email headers (sender, date, subject) are semi-structured; the free-text body is unstructured. Read what the question asks about.
• Treating JSON or XML as structured data. Fix: They use tags and flexible nesting without a fixed row-column schema, so classify them as semi-structured.
• Filling every missing value with the mean. Fix: Use the median when data is skewed or has outliers, and the mode for categorical fields. State your reason.
• Deleting every outlier automatically. Fix: Investigate first. A large genuine order is real information. Remove or correct only when it is an error.
• Calling a sales forecast descriptive analytics Fix: Ask whether the output is about the future. If yes, it is predictive.
• Treating predictive output as a certain result Fix: Say it is an estimate with error, based on past patterns that may change.

Exam tips

• Expect MCQs on definitions, the three components, the Vs of big data and which stage came when. Learn these as short lists.
• For written answers, a small comparison table with 4 to 5 points earns clear step marks.
• Write the evolution in time order. Examiners reward sequence and cause, not buzzwords.
• Always attach one Indian business example. It shows application, not just recall.
• There is no negative marking in the MCQ section, so attempt every question.
• Learn the stages in order and be ready to write them as a numbered list with one line each.
• If a case is given, use its business context in every stage instead of generic text.
• Mention that the process is iterative and that models need monitoring; many answers miss this.