Utilise various investigative approaches and methodologies for problem-solving in computing science research.
Lectures 1–3 covered identifying and validating a research problem. Once validated, the next intellectual task is to translate it into a clearly bounded scope and a set of well-articulated objectives. Without these, even a brilliant problem statement cannot be operationalised into a tractable study.
Three items here are examined repeatedly: the six dimensions of scope (Section 3.1), the delimitation versus limitation distinction (Section 3.3), and the ability to rewrite a woolly objective in SMART form (Section 4.2).
Be ready to construct the full Aim → Objective → Research Question → Hypothesis chain for an unseen problem in about ten minutes. That is a standard descriptive question.
1. Learning Objectives
- Distinguish clearly between the scope of a study and its objectives.
- Apply established frameworks such as SMART to formulate research objectives.
- Recognise delimitations and limitations in a computing research context.
- Critically evaluate a research proposal for the quality of its scope and objectives.
- Convert a high-level research problem into bounded, achievable objectives.
2. Key Concepts at a Glance
| Concept | Working definition |
|---|---|
| Research scope | The boundaries of a study — what will and will not be covered |
| Research aim (goal) | The broad, long-term intention of the research |
| Research objective | A specific, measurable, achievable outcome that contributes to the aim |
| Research question | A precise interrogative form of an objective |
| Hypothesis | A testable proposition that operationalises an objective in deductive research |
| Delimitation | An intentional boundary imposed by the researcher |
| Limitation | A constraint imposed externally, by data, time or resources |
| Operationalisation | Converting a conceptual objective into measurable indicators |
These are not synonyms. Conflating them is a frequent origin of methodological weakness.
3. Defining the Scope
3.1 The Six Dimensions of Scope
Scope is the perimeter inside which the entire inquiry is conducted. It answers: what exactly am I studying, where, when, on whom or on what, and under what conditions? In synopses, proposals and theses it is expressed through six dimensions:
- Domain or subject boundary — the subdomain of computer science: natural language processing, network security, human-computer interaction, distributed systems.
- Theoretical boundary — the theories, models or frameworks consulted: information retrieval theory, game theory, complexity theory.
- Population or sampling boundary — the datasets, users, software artefacts, hardware platforms or system events examined.
- Geographical boundary — the geographic or institutional context, for instance an IoT deployment in a single smart-campus testbed.
- Temporal boundary — the time horizon, for instance traffic logs collected between January and June 2024.
- Methodological boundary — the methods permitted, for instance only empirical benchmarks and no theoretical proofs.
3.2 The Funnel of Specificity
Each level narrows the inquiry. Scope is the next-to-last level; objectives are the smallest and most concrete formulations.
3.3 Delimitations versus Limitations
A delimitation is a boundary deliberately chosen by the researcher, often because exploring everything is impossible. “The study considers only supervised classifiers; unsupervised and semi-supervised methods are excluded.” “The proposed algorithm is benchmarked on English-language corpora only.”
A limitation is a boundary imposed externally. “The available GPU compute — a single A100 — restricts model size.” “Access to only one medical institution restricts patient diversity in the dataset.”
Delimitations are choices; limitations are constraints. Both must be acknowledged, but only delimitations are arguable.
3.4 Common Errors When Defining Scope
- Scope creep — gradual, uncontrolled expansion after the study has begun. In software engineering research this typically appears as adding new modules, datasets or comparison systems midway.
- Over-scoping — attempting to address too many sub-problems at once.
- Under-scoping — carving out a niche so narrow that the result has no generalisable or publishable contribution.
- Implicit scope — leaving boundaries unstated, so the reader must infer them.
4. Defining the Objectives
4.1 The Hierarchy: Aim, Objectives, Tasks
Aim and objectives should not be stated as the same sentence at the same level of abstraction. The aim is essentially non-testable; each objective is independently testable.
| Level | Example in a computing context |
|---|---|
| Aim | To improve the robustness of deep learning models against adversarial perturbations in image classification |
| Objective 1 | To design a novel adversarial training algorithm that reduces attack success rate by at least 15% on benchmark datasets |
| Objective 2 | To evaluate the trade-off between clean accuracy and adversarial robustness using standardised metrics such as robust accuracy |
| Objective 3 | To validate the proposed method across at least three representative attacks (FGSM, PGD, C&W) |
| Objective 4 | To compare the proposed method against three state-of-the-art baselines under identical experimental conditions |
4.2 The SMART Framework
- Specific — the objective identifies what will be done, to what, in which context.
- Measurable — a metric or criterion exists by which success can be judged. In computing this usually means accuracy, F1-score, AUC, latency in milliseconds, throughput in requests per second, energy in joules, or a qualitative rubric.
- Achievable — feasible given available datasets, compute, time and expertise.
- Relevant — aligned with the stated aim and contributing to the field.
- Time-bound — schedulable within the research window, often 6, 12 or 24 months.
4.3 Types of Objectives
By generality: general or primary objectives state the umbrella intent, usually in one statement; specific or secondary objectives are the operational breakdowns that collectively satisfy it.
By cognitive intent, drawing on Bloom's revised taxonomy. The cognitive level of an objective must match its actual deliverable:
| Cognitive level | Verb examples | Typical CS deliverable |
|---|---|---|
| Remember | identify, list, recall | Taxonomy of attacks |
| Understand | describe, summarise | Literature synthesis |
| Apply | implement, deploy | Prototype system |
| Analyse | compare, examine | Benchmark study |
| Evaluate | assess, justify | Comparative empirical study |
| Create | design, develop, propose | Novel algorithm or architecture |
A frequent methodological fault is to state an objective at the Create level but to conduct only an Analyse-level study.
4.4 Objectives, Research Questions and Hypotheses
In qualitative and mixed-methods research, each objective is often rephrased as a research question.
Objective: to evaluate the latency impact of edge offloading in mobile cloud computing.
Research question: what is the effect of edge offloading on request–response latency under varying network conditions?
In quantitative, positivist research each question is often rephrased as a falsifiable hypothesis:
Not every computing study uses hypotheses: design science and exploratory research often bypass the hypothesis form and proceed directly from objectives to artefacts.
5. How Scope and Objectives Constrain Each Other
Scope and objectives are not redundant; they are mutually constraining. A scope that is too wide makes the objectives unfalsifiable; objectives that are over-ambitious force the scope to expand beyond feasibility. A well-formulated study requires the scope to be just wide enough to make the objectives meaningful, and the objectives just narrow enough to fit the scope.
For each objective, ask: is there a documented, available and ethical way to gather the data required to satisfy this objective within the stated scope?
If the answer is no for any objective, either the scope or the objective must be revised.
6. Worked Examples
Initial, poorly formed: “AI is very useful. We want to use AI for healthcare.”
Defined scope: “This study focuses on deep learning models for chest X-ray classification, restricted to publicly available datasets (CheXpert and NIH ChestX-ray14), evaluated using AUROC and clinical sensitivity, on a single institutional GPU cluster, between the academic year 2024–2025.”
Defined objectives:
- To curate and standardise two benchmark chest X-ray datasets for binary pneumonia detection.
- To design a transfer-learning pipeline based on a pre-trained convolutional backbone.
- To evaluate the model using five-fold cross-validation against two state-of-the-art baselines.
- To interpret model decisions using Grad-CAM visualisations.
Each objective is SMART, internally consistent, and bounded by the stated scope.
Problem: insider threats remain a leading cause of data breaches in enterprises.
Scope: user behavioural logs from a single enterprise testbed; a six-month observation window; three observable channels (file access, email, login patterns); detection restricted to offline batch processing, not real-time streaming.
Objectives:
- To engineer a multi-view feature representation combining file-system, email and authentication logs.
- To train and tune an ensemble anomaly detector based on isolation forests and recurrent neural networks.
- To evaluate detection using precision, recall and F1 against a labelled ground-truth set.
- To estimate the operational cost, in false alarms per analyst-shift, of the proposed system.
A scan of the published literature on Android malware detection reveals a consistent pattern. Studies that clearly delimit themselves to a specific feature family — API-call sequences, or permissions — and a specific detection paradigm — static analysis only — consistently produce reproducible results. Studies that attempt to “cover all features and all detection methods” tend to underperform on any single benchmark, suffering from over-scoping.
Lesson: defining scope narrowly is often the path to stronger empirical evidence. This is counterintuitive but reliable.
6.1 Well-Defined versus Poorly-Defined Objectives
| Aspect | Poorly-defined objective | Well-defined objective |
|---|---|---|
| Specificity | “To study deep learning.” | “To fine-tune BERT variants for sentiment classification on three product-review datasets.” |
| Measurability | “To improve performance.” | “To achieve a macro-F1 improvement of at least 3 percentage points over the baseline.” |
| Achievability | “To solve general AI.” | “To reduce inference latency by 20% via INT8 quantisation on the same hardware.” |
| Relevance | “To use big data.” | “To mitigate class imbalance in real-time network intrusion detection.” |
| Time-bound | Implicit. | “Within an 8-month implementation window.” |
7. Common Errors When Defining Objectives
- Conflating activities with outcomes. “To perform a literature survey” is an activity; “to identify the three most influential research gaps in federated learning” is an outcome. Objectives must be outcomes.
- Multiple objectives in a single sentence. Compound objectives obscure which deliverable serves which evaluation. Split them.
- Objectives that exceed the scope. An objective requiring proprietary medical records when the scope restricts the study to public datasets is unachievable.
- Objectives that are not testable. “To deeply understand interpretability” cannot be measured and is therefore not a research objective in the strict sense.
- Objectives that contradict the stated aim. An aim of improving efficiency paired with an objective of improving accuracy, without acknowledging the trade-off, signals weak integration.
- Neglecting the time dimension. Without a schedule anchor, objectives become aspirations rather than commitments.
8. Summary
- Scope establishes the boundaries of the study — domain, theory, population, geography, time and methodology. Delimitations are chosen; limitations are imposed.
- Objectives are the measurable outcomes through which the aim is realised. They should follow SMART and align with a clear cognitive level.
- A sound design requires mutual consistency between scope and objectives: each objective must be satisfiable within the declared boundaries.
- In computing research, where artefacts — algorithms, systems, datasets — often constitute the contribution, objectives typically lead to one or more artefacts, each with an evaluation plan.
- Writing tight scope statements and SMART objectives is itself a strong indicator of research quality, and among the first things examiners and reviewers assess.
Defining scope and objectives is therefore not a bureaucratic preliminary. It is the act that converts intellectual curiosity into executable research.
9. Exam Preparation
- Memorise the dimensions of scope. Examiners ask students to enumerate them from memory.
- Be fluent with SMART. Convert poorly formed objectives into SMART form during revision; this builds speed.
- Practise the Aim → Objective → RQ → Hypothesis chain. Be ready to construct it for any given problem in ten minutes.
- Distinguish delimitations from limitations — a recurring one- and two-mark question.
- Know Bloom's verbs. Pair each CS artefact — model, benchmark, framework — with its appropriate cognitive level.
9.1 Short-Answer Questions
- Differentiate between delimitation and limitation in a computing research context. Provide one example of each from a study on cloud security. 3 marks
- List the five elements of the SMART framework and rewrite the following objective in SMART form: “To study machine learning algorithms.” 3 marks
- Why must the cognitive level of an objective match the actual deliverable of the study? Illustrate with a computing example. 3 marks
9.2 Descriptive Questions
- Take the following research problem and, in approximately 200 words, define the scope and write three SMART objectives: “Online learning platforms face challenges in student dropout prediction.” 10 marks
- Discuss with examples the common errors researchers make while defining research scope. How does scope creep manifest in software-engineering research projects? 8 marks
- Convert the following aim into a hierarchical structure — one aim, four specific objectives, corresponding research questions, and two hypotheses where applicable: “To improve the privacy of federated learning systems.” 10 marks
9.3 Higher-Order Question
A student submits the following objective in a research proposal: “To study various deep learning models comprehensively for image classification to see which is best.”
Identify at least five methodological weaknesses in this objective. Rewrite it into three well-formed SMART objectives consistent with a defined scope in image classification research. 15 marks