FRM Exam Part I · Machine-Learning Methods

Machine Learning Basics and Types of Learning for FRM Part I

Updated 11 October 2026 · Fact-checked

Machine learning lets algorithms find patterns in data and improve predictions without hand-written rules. Supervised learning uses labeled outcomes, unsupervised learning finds structure in unlabeled data, and reinforcement learning learns actions through rewards. To answer exam questions, check whether labels exist, then match the goal to the learning type.

Understand Overview of Machine Learning and Types of Learning

Machine learning (ML) is a set of methods where a model learns patterns from data and uses them to predict or to group observations. You do not write the rule yourself. The algorithm finds it.

Traditional statistics usually starts with a theory and a model form, such as a linear regression with chosen variables. It focuses on inference: testing hypotheses, estimating coefficients and judging significance. ML focuses on prediction. It can handle many variables, nonlinear relationships and interactions, often with less structure imposed up front. The price is lower interpretability and a higher risk of overfitting.

Supervised learning uses data with a known outcome, called a label or target. The model learns the link between inputs (features) and the target. If the target is a number, such as a loss amount, it is regression. If the target is a category, such as default or no default, it is classification. Typical finance uses: credit scoring, default prediction, fraud detection, forecasting returns or volatility.

Unsupervised learning has no labels. The model looks for structure in the features alone. Main tasks are clustering (grouping similar borrowers or trades) and dimension reduction (compressing many correlated variables into a few, as in principal components analysis). Finance uses: customer segmentation, spotting unusual transactions, summarizing yield curve moves.

Reinforcement learning has an agent that takes actions in an environment and receives rewards or penalties. It learns a policy that maximizes cumulative reward over time. There is no fixed labeled answer for each step. Finance uses: optimal trade execution, dynamic hedging, portfolio allocation over time. Remember the key idea: the action changes the next state, so decisions are sequential.

Key formulas to remember

Supervised learning: Inputs (features) + known labels → learn a mapping to predict the label; Numeric label = regression. Categorical label = classification.
Unsupervised learning: Inputs (features) only, no labels → find clusters or lower-dimensional structure; Tasks: clustering and dimension reduction.
Reinforcement learning: Agent takes action in a state → receives reward → updates policy to maximize cumulative reward; Sequential decisions; feedback is a reward, not a correct label.
Prediction versus inference: ML: emphasis on out-of-sample prediction. Traditional statistics: emphasis on estimating and testing parameters; A guide to the usual emphasis, not a strict rule. Many methods serve both.

How to solve Overview of Machine Learning and Types of Learning questions

Use this sequence for any question that asks you to classify a method, a problem or an application.

1Read the scenario and find the goal: predict a value, assign a category, find groups, reduce variables, or choose actions over time.
2Ask whether the data contain a known outcome for each observation. If yes, think supervised.
3If supervised, check the outcome type: number means regression, category means classification.
4If there is no outcome and the aim is to discover structure, think unsupervised: clustering or dimension reduction.
5If the scenario has an agent, repeated decisions and rewards, choose reinforcement learning.
6If the question contrasts ML with traditional statistics, think prediction and flexibility versus inference and interpretability.
7Check the options for traps such as swapping labels or calling clustering supervised, then choose the one that fits all facts.

Quickest way: Three-question label test

When to use it: Use it on any multiple-choice question that asks which type of learning applies.

Is there a labeled outcome? Yes: supervised. Then number = regression, category = classification.
No label, goal is groups or fewer variables? Unsupervised.
Rewards after actions, decisions in sequence? Reinforcement.

Common mistakes in Overview of Machine Learning and Types of Learning

Calling clustering a supervised method.
Groups sound like categories, so students assume labels exist.
Fix: In clustering the groups are discovered, not given. If no labels were supplied in training, it is unsupervised.
Treating default prediction as unsupervised because it is about risk.
Students focus on the topic instead of the data structure.
Fix: If past loans are tagged default or not default, it is supervised classification.
Confusing regression with classification in supervised learning.
Logistic regression has 'regression' in its name.
Fix: Look at the output. A category or probability of a class means classification, even for logistic regression.
Thinking reinforcement learning needs labeled correct answers.
Students assume all training needs examples of right answers.
Fix: Reinforcement learning uses reward feedback from actions, often delayed, not a label per step.
Saying ML is always better than traditional statistics.
ML is often described as more powerful.
Fix: ML can predict well but may be hard to interpret and can overfit. Statistics suits inference and small data.
Describing dimension reduction as a way to predict a target.
Students see it used before regression and mix up the roles.
Fix: Dimension reduction summarizes features without a target. It can feed a supervised model later, but itself is unsupervised.

Worked examples

Example 1

A bank has 10 years of loan records, each marked as defaulted or not. It wants a model to estimate whether a new applicant will default. Which type of learning and which task is this?

Show the solution

Goal: predict an outcome for new applicants.
Labels exist: each past loan is marked default or not default.
Labels exist, so the learning is supervised.
The outcome is a category with two values, so the task is classification, not regression.

Answer: Supervised learning, classification.

Example 2

A risk team has transaction data for thousands of clients with no labels. It wants to find groups of clients with similar behavior and then review small unusual groups. Which learning type applies, and could it use reward feedback?

Show the solution

There are no labels, so this is not supervised.
The goal is to discover groups, which is clustering.
Clustering finds structure in features only, so the type is unsupervised.
No agent takes actions and receives rewards, so reinforcement learning does not apply.

Answer: Unsupervised learning (clustering); reward feedback is not part of it.

Exam tips

Decide on labels first. Most questions on this topic are answered by that single check.
Watch for key words: 'labeled', 'target' mean supervised; 'segment', 'group', 'compress' mean unsupervised; 'agent', 'reward', 'policy' mean reinforcement.
Expect finance links: credit scoring and fraud as supervised, customer segmentation as unsupervised, execution and hedging as reinforcement.
In contrast questions, link ML to prediction and flexibility, and traditional statistics to inference and interpretability, without saying one always wins.

Practice questions from Machine-Learning Methods

Overview of Machine Learning and Types of Learning in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Overview of Machine Learning and Types of Learning: frequently asked questions

What is the difference between supervised and unsupervised learning?

Supervised learning trains on data with known outcomes and learns to predict them. Unsupervised learning has no outcomes and looks for patterns such as clusters or main components in the features.

How is reinforcement learning used in finance?

It suits sequential decisions where actions affect later results, such as trade execution, dynamic hedging and portfolio rebalancing. The agent learns a policy that maximizes cumulative reward.

Is logistic regression machine learning or statistics?

It is both. It comes from statistics but is used as a supervised classification method in ML, for example to model the chance of default.

How is machine learning different from traditional statistics for the FRM?

Traditional statistics usually stresses inference with a chosen model form. ML stresses prediction, handles many variables and nonlinear patterns, and needs care about overfitting and interpretability.