Predictive Analytics (AD-702 (D)) Unit 1 Notes

How unit 1 is examined

This unit covers what predictive analytics is, the three kinds of analytics and models (predictive, descriptive, decision), the analytical techniques used, and how data is classified, prepared and explored; none of the ten topics was asked in the supplied papers, so each is kept short but complete enough to answer if it appears.

Introduction to predictive analytics

Not asked since 2022

Definition. Predictive analytics uses historical data, statistics and machine learning to estimate the probability or value of a future outcome.

Key points.

It learns a pattern from past records and applies it to new cases, for example predicting customer churn, loan default or next month's sales.
The output is a score, probability or forecast, never a certainty, so it supports a decision rather than replacing it.
The usual flow is: define the goal, collect data, prepare it, build a model, validate it on unseen data and deploy it.
Its accuracy depends on data quality and on the future resembling the past.

Business analytics: types, applications

Definition. Business analytics is the use of data, statistics and quantitative methods to support and improve business decisions.

Key points.

Descriptive analytics answers "what happened" using reports, dashboards and summaries.
Predictive analytics answers "what will happen" using models and forecasts.
Prescriptive analytics answers "what should we do" using optimization and simulation.
Applications include retail demand forecasting, credit scoring, fraud detection, targeted marketing, churn prevention and supply-chain planning.

Predictive models

Definition. A predictive model maps input variables to a target variable so that the target can be estimated for unseen data.

Key points.

Regression models predict a numeric target such as sales, for example $\hat{y}=\beta_0+\beta_1x_1+\dots+\beta_kx_k$.
Classification models predict a category such as fraud or not fraud, and time-series models forecast values over time.
They are trained on labelled historical data, which makes them supervised.
They must be judged on held-out test data, not on the training data.

Descriptive models

Definition. A descriptive model summarizes and finds structure in existing data without predicting a target variable.

Key points.

Clustering groups similar customers into segments, which is called customer profiling.
Association rules find items that occur together, such as bread and butter in market-basket analysis.
They are usually unsupervised because no target label is used.
Their output explains the present data and often becomes an input to later predictive models.

Decision models

Definition. A decision model uses objectives, constraints and predictions to recommend the best action.

Key points.

It is the prescriptive step and typically uses optimization, such as linear programming, or simulation.
Its inputs are decision variables, an objective (maximize profit or minimize cost) and constraints such as budget or capacity.
It often consumes predictions, for example forecast demand, to choose stock levels or prices.
Its output is a recommended action, not a forecast.

Analytical techniques

Definition. Analytical techniques are the statistical and data-mining methods used to extract patterns and build models from data.

Key points.

Statistical techniques include regression, hypothesis testing, correlation and time-series analysis.
Data-mining techniques include classification, clustering, association rules and decision trees.
Machine-learning methods such as neural networks and ensembles learn complex non-linear patterns from data.
The technique is chosen by the goal (predict, group or optimize) and by the type of data available.

Data types and associated techniques

Definition. Data is structured, semi-structured or unstructured, and each type needs different techniques.

Key points.

Structured data sits in tables with fixed fields and suits SQL, regression and classification.
Semi-structured data such as JSON or XML has tags but no fixed schema and needs parsing before analysis.
Unstructured data such as text, images and audio needs text mining, computer vision or speech processing.
Variables are also numeric (continuous or discrete) or categorical (nominal or ordinal), and this decides which statistics and encodings are valid.

Complexities of data

Definition. Data complexity is the set of properties, such as large volume, wide variety and poor quality, that make data hard to analyze.

Key points.

Volume, velocity and variety of big data strain storage and processing.
Missing values, noise, outliers and duplicates reduce the accuracy of models.
Data from many sources has inconsistent formats and units and needs integration.
High dimensionality and imbalanced classes make models overfit or ignore rare events such as fraud.

Data preparation, pre-processing

Definition. Data preparation converts raw data into a clean, consistent form suitable for modelling.

Key points.

Cleaning fills or removes missing values, corrects errors and removes duplicates and outliers.
Integration merges data from several sources into one consistent set.
Transformation applies scaling, normalization $x'=\frac{x-x_{min}}{x_{max}-x_{min}}$, and encoding of categories as numbers.
Reduction lowers dimensions or rows through feature selection or sampling, and the whole step usually takes most of the project time.

Exploratory data analysis

Definition. Exploratory data analysis (EDA) is the visual and numerical examination of data to understand its structure, spot anomalies and form hypotheses before modelling.

Key points.

Summary statistics such as mean, median, standard deviation and quartiles describe each variable.
Histograms, box plots and scatter plots show distribution, outliers and relationships.
A correlation matrix reveals which variables move together and which are redundant.
Findings guide cleaning, feature choice and model selection.

Last-minute revision

Predictive analytics estimates future outcomes from historical data using statistics and machine learning.
Its flow: define goal, collect, prepare, model, validate, deploy.
Three types of analytics: descriptive (what happened), predictive (what will happen), prescriptive (what to do).
Predictive models are supervised: regression for numbers, classification for categories.
Regression form: $\hat{y}=\beta_0+\beta_1x_1+\dots+\beta_kx_k$.
Descriptive models (clustering, association rules) are usually unsupervised.
Decision models use optimization or simulation with an objective and constraints.
Data is structured, semi-structured (JSON, XML) or unstructured (text, images, audio).
Variables are numeric (continuous, discrete) or categorical (nominal, ordinal).
Complexities: volume, variety, velocity, missing values, noise, outliers, imbalance.
Preparation steps: clean, integrate, transform, reduce.
Min-max normalization is $x'=\frac{x-x_{min}}{x_{max}-x_{min}}$, giving values between 0 and 1.
EDA uses summary statistics, plots and correlation before any modelling.

Memory hooks

D-P-P: Describe, Predict, Prescribe.
CITR for preparation: Clean, Integrate, Transform, Reduce.
Regression = number, classification = label.
Descriptive = no target, predictive = has target, decision = has constraints.
SSU for data: Structured, Semi-structured, Unstructured.
EDA = look before you model.

Coverage checklist

Introduction to predictive analytics: definition, flow and limits (no past questions).
Business analytics: types, applications: descriptive, predictive, prescriptive and uses (no past questions).
Predictive models: regression and classification (no past questions).
Descriptive models: clustering and association rules (no past questions).
Decision models: optimization and simulation (no past questions).
Analytical techniques: statistical, data mining and machine learning (no past questions).
Data types and associated techniques: structured, semi-structured, unstructured, numeric, categorical (no past questions).
Complexities of data: volume, quality, integration, imbalance (no past questions).
Data preparation, pre-processing: cleaning, integration, transformation, reduction (no past questions).
Exploratory data analysis: summary statistics, plots, correlation (no past questions).

Account

Predictive Analytics (AD-702 (D)) - Unit 1 Short Notes

How unit 1 is examined

Introduction to predictive analytics

Business analytics: types, applications

Predictive models

Descriptive models

Decision models

Analytical techniques

Data types and associated techniques

Complexities of data

Data preparation, pre-processing

Exploratory data analysis

Last-minute revision

Memory hooks

Coverage checklist