Reducing Combinations with Equivalence Partitioning and Classification Trees

Reducing Combinations with Equivalence Partitioning and Classification Trees (ISTQB CTAL-TA)

Overview
In the ISTQB Advanced Level Test Analyst (CTAL-TA) syllabus, combinatorial testing is one of the data-based test techniques. It cannot work well when parameters have huge or infinite value ranges, because the number of combinations grows very quickly. The standard solution is to reduce each parameter to a small set of representative values first, using equivalence partitioning (EP) and often a classification tree. This guide explains why the reduction step matters, what it is, how it works, and how to handle related exam questions.

1. Why It Is Important
Combinatorial explosion: The number of possible combinations is the product of the number of values of every parameter. Five parameters with 10 values each give 10 x 10 x 10 x 10 x 10 = 100,000 combinations. If one parameter is a numeric field such as 'amount' (0.01 to 1,000,000.00), the total is effectively unlimited.
Combinatorial techniques need finite, small value sets: Pairwise testing, each choice testing and base choice testing all assume that each parameter has a small, discrete list of values. EP turns large or continuous domains into a few classes, so these techniques become practical.
Efficiency without losing much defect detection: EP assumes the system treats every value in a partition the same way. One representative per partition gives nearly the same defect-finding power as testing every value, at a tiny fraction of the cost.
Structure and communication: A classification tree shows the test-relevant aspects and their classes in one diagram. Stakeholders can review it, gaps become visible, and test cases can be traced back to it.
Handling constraints: Reduction also means removing infeasible or meaningless combinations. For example, 'Safari' together with 'Windows 11' may not be supported. This avoids wasted tests.

2. What It Is
Equivalence partitioning: A black-box technique that divides the input or output domain of a parameter into partitions (classes). The test object is expected to process all members of a partition in the same way.

• Valid partitions contain values the system should accept.
• Invalid partitions contain values it should reject.
• Partitions must not overlap. Together they should cover the whole domain.

Classification Tree Method (CTM): A graphical technique that applies partitioning systematically to every test-relevant aspect of the test object.

• Root: the test object, for example 'Car insurance premium calculation'.
• Classifications: the test-relevant aspects or parameters, for example Driver age, Vehicle type, Region.
• Classes: the equivalence partitions of each classification, for example 18-24, 25-64, 65+. Classes of one classification are disjoint and complete.
• Hierarchical refinement: a class can itself be refined into sub-classifications. For example, Vehicle type = Car can be refined by Engine size, which does not apply to Motorbike. This reduces combinations further, because sub-classifications only combine where they are relevant.
• Combination table: the grid under the tree. Each row is a test case that selects exactly one class from each relevant classification.

Combinatorial testing: Techniques that select a subset of combinations of parameter values according to a coverage criterion. Common criteria:

• Each choice: every value of every parameter appears at least once.
• Base choice: start from a base test, then vary one parameter at a time.
• Pairwise (2-wise): every pair of values from any two parameters appears together at least once.
• n-wise: the same idea extended to every combination of n parameters.
• All combinations: the full Cartesian product.

The reduction step is what feeds the combinatorial step. The classes become the 'values' that the combinatorial technique combines.

3. How It Works: Step by Step
Step 1: Analyse the test basis. Identify the test object and all test-relevant aspects: inputs, configuration settings, environment factors and states.
Step 2: Define classifications. List each aspect as a classification or parameter. Leave out aspects that do not affect behaviour.
Step 3: Partition each classification into classes (EP). Use specifications, business rules and boundaries. For example, Age from 0 to 120 becomes the invalid class <18 and the valid classes 18-24, 25-64 and 65-120. That is 4 classes instead of 121 values.
Step 4: Refine hierarchically where needed. Add sub-classifications only under the classes where they matter.
Step 5: Separate invalid classes. Invalid classes are normally tested one per test case, combined with valid classes for all other parameters. This prevents defect masking, where one invalid value triggers an error that hides the effect of another. Usually only valid classes go into the combinatorial step.
Step 6: Identify constraints. Mark impossible or forbidden combinations, such as 'Motorbike' combined with 'Engine size > 3000cc' if that is a business rule.
Step 7: Choose the coverage level and generate combinations. Pick the level based on risk: each choice for low risk, pairwise for medium risk, higher n-wise for high risk. Use a tool such as PICT or a CTM tool, or the combination table.
Step 8: Select concrete representative values. For each class in each test case, choose an actual value. BVA can be added for boundary values, often as separate tests.

Worked Example
Online insurance quote with these parameters:

• Driver age: integer 0-120
• Vehicle type: Car, Motorbike, Van
• Region: North, South, East, West, Central
• Payment: Monthly, Annual

Before reduction: 121 x 3 x 5 x 2 = 3,630 combinations.
After EP: Age is split into the invalid class <18 (tested separately) and the valid classes 18-24, 25-64 and 65+.

• All combinations of valid classes: 3 x 3 x 5 x 2 = 90.
• Each choice: at least the largest number of classes in any classification, so 5 tests (Region has 5).
• Pairwise: at least the product of the two largest class counts, so 5 x 3 = 15 tests. Tools typically achieve close to this.
• Plus 1 separate test for the invalid age class.

The test set drops from thousands to roughly 6 to 16 tests, depending on the coverage criterion.

Useful Formulas

• Full combinations = product of the number of classes per parameter (minus combinations excluded by constraints).
• Minimum for each choice = the largest number of classes in any single parameter.
• Lower bound for pairwise = the product of the two largest class counts.
• Minimum tests for invalid classes, if tested one at a time = the number of invalid classes.

4. Benefits and Limitations
Benefits: large reduction in test count, systematic coverage, visual communication, traceability, and easy maintenance when requirements change.
Limitations:

• Quality depends on correct partitions. A wrong partition means missed defects.
• Representatives may miss boundary defects, so combine with BVA.
• Interactions involving more than 2 parameters may be missed by pairwise testing.
• Constraints must be modelled carefully.
• Building the tree requires domain knowledge.

5. Exam Tips: Answering Questions on Reducing Combinations with Equivalence Partitioning and Classification Trees
Tip 1: Know the purpose. If a question asks why EP or CTM is applied before combinatorial testing, the answer is to reduce the number of values per parameter to a manageable set of representative classes. This keeps the number of combinations practical. Reject distractors such as 'to achieve full path coverage' or 'to eliminate the need for boundary testing'.
Tip 2: Master the terminology. Classification = parameter or aspect. Class = equivalence partition of that aspect. Root = test object. Each test case picks exactly one class per relevant classification. Watch for distractors that swap 'class' and 'classification'.
Tip 3: Do the arithmetic carefully.

• For 'how many combinations', multiply the class counts.
• For 'minimum tests for each choice', take the maximum class count.
• For 'minimum tests for pairwise', the lower bound is the product of the two largest class counts.

Read whether invalid classes are included. If the question says invalid classes are tested separately, add them as single extra tests and do not multiply them in.
Tip 4: Invalid partitions are tested one at a time. If an option combines several invalid values in one test, it is usually wrong, because of defect masking.
Tip 5: Look for constraints. If the scenario says some combinations are impossible, subtract them or exclude them. Questions often hide a constraint in the scenario text.
Tip 6: Hierarchical refinement reduces combinations. Sub-classifications apply only under their parent class. Do not multiply them across all branches.
Tip 7: Check partition quality. In scenario questions, verify that partitions are disjoint, complete and based on behaviour. Overlapping ranges such as 18-25 and 25-64 indicate an error. Missing ranges indicate incomplete coverage.
Tip 8: Match the coverage level to risk. Higher risk calls for a stronger criterion (pairwise or 3-wise). Low risk or limited time may justify each choice. If the question gives risk information, use it.
Tip 9: Remember complementary techniques. EP plus CTM reduces combinations. BVA adds boundary values. A strong answer may mention adding BVA tests for boundaries of the partitions.
Tip 10: Eliminate extreme answers. Options claiming that combinatorial testing 'guarantees finding all interaction defects' or that EP 'removes the need for any other technique' are typically wrong.
Tip 11: Time management. For calculation questions, first list each parameter with its class count. Then decide valid versus invalid. Then apply the formula. This avoids careless errors under exam pressure.

Quick Summary
Identify the aspects, partition them with EP, and organise them in a classification tree. Refine hierarchically, separate invalid classes, and model constraints. Then apply a combinatorial criterion to the valid classes. The result is a small, systematic and risk-appropriate set of test cases instead of an unmanageable combinatorial explosion.