Machine Learning Problem Definition

Think of machine learning as a workflow of connected decisions, not only as model construction.

Programming

From Goal to Model

Machine learning is easier to reason about when you treat it as a workflow of connected decisions rather than as one act of building a model. The work begins with a real-world problem: what should the system help accomplish, what information will it receive, and what target output should it produce? From there, the team prepares data, decides how success will be measured, develops a model, and checks whether the result solves the broader problem rather than merely fitting the available training data.

What happens next as a machine learning problem moves from defining the goal through data preparation, feature engineering, model training, evaluation, and revision?

Defining the Target

Before evaluating models, define the real-world problem, the inputs available to the system, and the target output it should produce. The target gives the task a clear destination. The inputs describe the information from which the system is expected to produce that target. Without these decisions, it is difficult to know what data to prepare or what a model's result is supposed to mean.

Worked example

A New Machine Learning Scenario

A team wants to create a machine learning solution for a real-world task and has collected raw data, but has not yet stated the target output, selected a measure of success, prepared the data, or considered overfitting.

1
State the problem
Describe the real-world task the solution is intended to support before focusing on model construction.
2
Identify inputs
Specify which information will be available for the machine learning system.
3
Identify the target
State the output the system should produce. This turns a broad task into a defined machine learning problem.
4
Continue the workflow
After the target is defined, choose a measure of success, prepare the raw data, create useful features, and plan how to evaluate the model.

Result

The team has changed an unfinished idea into a sequence of decisions that can be prepared, modeled, and evaluated.

How do the problem goal, target, features, evaluation measure, and modeling decisions connect in a new scenario?

Preparing the Data

Raw data does not move directly into a model as one undifferentiated collection. Preparation can involve cleaning the available data, transforming it, preparing features, and separating the data into training and evaluation sets. These stages create a usable path from raw information to a model. Feature engineering belongs inside this path: it is the preparation of useful input features from the available data.

How does raw data change as it is collected, cleaned, split, transformed into features, and passed to a model?

How are useful input features derived from prepared data before they reach the model?

Choosing Success

A model cannot be judged without a measure of success. This measure should be chosen as part of defining the machine learning problem, before results are evaluated, rather than added after a model already exists. The measure determines what the project treats as a successful result. It therefore connects the real-world objective to the way model performance is judged.

How does the chosen evaluation measure connect the real-world objective to the way a model is judged?

Recognizing Overfitting

After data preparation and evaluation planning, model development must still confront overfitting. Overfitting occurs when a model fits the training data too closely. A close fit to the data used for training is not automatically the same as solving the broader machine learning problem. The workflow makes this danger visible by requiring evaluation beyond the training data.

How can a model perform well on training data while performing poorly on new, unseen data?

In the real world

Imagine a model that matches the training data very closely. That result may look encouraging, but the workflow still asks whether the model performs well when judged with the evaluation data and the chosen measure of success. If its evaluation result is weak, the model has fitted the training data too closely instead of addressing the broader problem.

Mistakes in Workflow Order

Common mistake
Starting with model construction before defining the real-world problem
What it looks like
A team begins developing a model while the target output is still unstated.
Why it fails
The team has not yet decided what the model should produce or which inputs belong to the task.
Do this instead
Define the problem, inputs, and target output before evaluating models.
Common mistake
Treating data preparation as a single isolated action
What it looks like
A team says the data is prepared without considering cleaning, transformation, feature preparation, or separation into training and evaluation sets.
Why it fails
Raw information needs a usable path to the model, and preparation can involve several connected stages.
Do this instead
Consider the full preparation path from raw data through cleaning, transformation, feature preparation, and separation.
Common mistake
Choosing the measure of success after seeing model results
What it looks like
A team builds a model first and only then decides how to judge it.
Why it fails
The evaluation measure may not reflect the original real-world objective.
Do this instead
Choose a measure of success as part of defining the problem.
Common mistake
Assuming a strong training fit proves the broader problem is solved
What it looks like
A model fits the training data closely, so the team stops evaluating.
Why it fails
A model can overfit the training data without addressing the broader machine learning problem.
Do this instead
Use evaluation data and the chosen measure of success to check the result.

Best practice

At each stage, ask what decision the previous stage enables and what decision the next stage requires. The problem definition guides the evaluation plan; the evaluation plan helps define what data and features matter; prepared features support model development; and evaluation reveals whether the model has overfit the training data.

Workflow Practice

MEDIUM

A team has collected raw data for a new machine learning solution. The team has not stated the target output, chosen a measure of success, prepared the data, or considered overfitting. Put the next decisions in a sensible workflow order and explain why each decision belongs where you place it.

Hints

Begin with the real-world problem, its inputs, and its target output.
Place the measure of success before judging model results.
Include cleaning, transformation, feature preparation, and separation into training and evaluation sets.
Explain why a close fit to training data is not enough.

Your attempt (not saved)

Worked example

A Complete Reasoning Path

Apply the workflow to the unfinished machine learning scenario.

1
1. Define the task
State the real-world problem and identify the inputs and target output.
2
2. Choose success
Choose the measure that will determine whether the model's result counts as successful.
3
3. Prepare the data
Move from raw data through cleaning, transformation, feature preparation, and separation into training and evaluation sets.
4
4. Develop the model
Use the prepared features and training data to develop a model.
5
5. Check for overfitting
Compare the model's behavior on training data with its evaluation result. A close fit to training data alone does not establish that the broader problem has been solved.
6
6. Revise when needed
Use the evaluation result to reconsider the connected decisions in the workflow rather than treating model construction as the only important step.

Result

The workflow connects definition, evaluation, data preparation, feature engineering, model development, and overfitting checks into one sequence of decisions.

Key Takeaways

1
Machine learning is a workflow of connected decisions, not only the construction of a model.
2
Define the real-world problem, inputs, and target output before evaluating models.
3
Prepare raw data through cleaning, transformation, feature preparation, and separation into training and evaluation sets.
4
Choose a measure of success before judging results because it defines what the project treats as successful.
5
Overfitting means fitting the training data too closely; a strong training fit does not by itself solve the broader problem.

At a glance

Define the real-world task, inputs, and target output before focusing on model evaluation.
Treat cleaning, transformation, feature preparation, and separation into training and evaluation sets as connected preparation stages.
Choose the measure of success as part of defining the problem, not as an afterthought.
Recognize overfitting when a model fits training data too closely and performs inadequately as a broader solution.
Use the workflow to connect problem definition, evaluation, features, model development, and revision.

Related concepts

More concepts starting with M