The Four Types of Analytics
Think of these as a ladder — each level builds on the one before it, and each one adds more value to business decisions.
Descriptive Analytics — this is your starting point. It summarizes what already happened using historical data. Think dashboards, reports, and averages.
A hospital creates a monthly report showing that emergency room visits increased 15% last quarter compared to the same period last year. That's descriptive analytics — it tells you what happened, but not why.
Diagnostic Analytics — once you know what happened, you naturally ask why. This involves drilling into patterns, correlations, and root causes.
The hospital discovers that the ER spike was concentrated on weekends and correlated with flu season starting earlier than usual. They used drill-downs, filtering, and trend comparisons to find the pattern.
Common techniques here include data exploration, correlation analysis, and visual drill-downs in tools like SAS Visual Analytics or Tableau.
Predictive Analytics — this is where our course lives! Predictive analytics uses historical data to estimate what's likely to happen in the future. The key word is probability, not certainty.
The hospital builds a model to predict which patients admitted today are most likely to be readmitted within 30 days. This lets them intervene early — maybe with extra follow-up calls or adjusted discharge plans.
This is the domain of linear regression, logistic regression, decision trees, random forests, and neural networks. We've already worked with linear regression, and we're about to add logistic regression and decision trees to your toolkit.
Prescriptive Analytics — the most advanced level. It doesn't just predict what will happen; it recommends what to do about it. This involves optimization, simulation, and decision modeling.
The hospital's system predicts a surge in ER visits next weekend and automatically recommends scheduling 3 additional nurses for the Saturday night shift, based on cost constraints and patient safety targets.
While we focus primarily on predictive analytics in this course, understanding prescriptive analytics helps you see where the field is heading — and where the biggest business value often lies.
Think of them as a progression — each builds on the last:
| Type | Key Question | Business Value |
|---|---|---|
| Descriptive | What happened? | Creates a shared understanding of the past |
| Diagnostic | Why did it happen? | Identifies root causes and patterns |
| Predictive | What will happen? | Enables proactive decisions |
| Prescriptive | What should we do? | Optimizes actions and resources |
Supervised vs. Unsupervised Learning
The biggest fork in the road when choosing a model: do you have a target variable, or are you exploring the unknown?
Supervised learning means you have a target variable — a specific outcome you're trying to predict or classify. The model learns from historical examples where you already know the answer.
Think of it like teaching with an answer key. You show the model thousands of labeled examples: "This customer churned. This one didn't. This one churned. This one didn't." The model finds patterns in the inputs that predict the target.
Ask yourself: "Am I predicting a specific outcome?" If yes → supervised. The target could be categorical (yes/no, fraud/not fraud) or continuous (revenue amount, temperature).
Unsupervised learning means there's no target variable. You're not predicting anything — you're discovering hidden structure in the data.
It's like sorting a pile of mixed laundry without labels telling you which drawer each item goes in. The algorithm finds natural groupings based on similarity.
Ask yourself: "Am I looking for hidden groups or patterns, without a specific outcome in mind?" If yes → unsupervised.
Both are supervised, but they answer different types of questions:
| Classification | Regression | |
|---|---|---|
| Target type | Categorical (groups, labels) | Continuous (numbers) |
| Question | "Which category?" | "How much?" or "How many?" |
| Example | Will this customer churn? (Yes/No) | How much will this customer spend? |
| Output | A label + probability | A number |
If you can count the possible answers on your fingers → classification.
If the answer could be any number → regression.
Where Do Common Models Fit?
Now let's map the models you'll encounter to their categories. This is the mental map that helps you pick the right tool for the job.
Here's a map of the models you'll see in this course and where they land:
| Model | Type | What It Does |
|---|---|---|
| Linear Regression | Supervised | Predicts a continuous number (e.g., sales revenue) |
| Logistic Regression | Supervised | Predicts the probability of a category (e.g., churn yes/no) |
| Decision Trees | Supervised | Classifies or predicts by splitting data into branches |
| Random Forest | Supervised | Combines many decision trees for stronger predictions |
| Neural Networks | Supervised | Finds complex, non-linear patterns in data |
| k-Means Clustering | Unsupervised | Groups similar observations together (e.g., customer segments) |
| Association Rules | Unsupervised | Finds items that occur together (e.g., market basket analysis) |
Great question — and there's a clear reason for the sequence.
With linear regression, you learned to predict how much — a continuous number. But many of the most important business problems aren't about "how much." They're about "will it or won't it?"
Will this customer cancel? Will this loan default? Will this patient be readmitted? Is this transaction fraudulent? Will this applicant accept our offer?
Logistic regression gives you the probability of a yes/no outcome. It's elegant, interpretable, and widely used in banking, healthcare, and marketing.
Decision trees take a completely different approach — they split data into groups using a series of if-then rules. They're visual, intuitive, and easy to explain to business stakeholders who don't speak "statistics."
Together, these two techniques give you two powerful lenses for the same type of problem: classification.
Check Your Understanding
Five quick scenarios — apply what you just reviewed. No grades, just practice.
Coming Up Next
You've refreshed the foundation. Now we're adding two new tools to your analytics toolkit: logistic regression for estimating probabilities and decision trees for building visual, rule-based classifiers. Both answer the same core question — "which category does this belong to?" — but they get there in very different ways.