How to Choose a Research Topic in Mathematics, Statistics or Data Science

Choosing a research topic is the most important decision of your BS, MS or PhD thesis, and it is also the one most students rush. A good topic keeps you motivated for months, fits your skills, can be finished on time and produces a result your examiners will respect. This guide gives you a step-by-step method, a scoring tool for comparing ideas and two worked feasibility calculations you can adapt for your own proposal.

What Makes a Good Research Topic?

Strong topics share five features:

1. Interest: you will live with it for a year or more, so curiosity matters.
2. Gap: it answers a question that published work has not fully settled.
3. Feasibility: the data, software, supervision and mathematical background are within reach.
4. Scope: narrow enough to finish, broad enough to say something meaningful.
5. Value: a result that matters to someone, whether other researchers, a hospital, a bank or a policy-maker.

A topic such as "Machine learning in healthcare" fails the scope test. "Comparing random forests and logistic regression for predicting gestational diabetes using hospital records from Punjab" passes: it names the methods, the outcome, the data and the setting.

Step 1: Start from What You Know

List the courses you enjoyed and did well in. In mathematics these might be numerical analysis, graph theory or differential equations; in statistics, regression, survival analysis or Bayesian methods; in data science, deep learning, natural language processing or time series forecasting. Your strongest course is usually the safest foundation for a thesis.

Step 2: Read Recent Literature Strategically

Search Google Scholar and your university's digital library for review articles published in the last five years on your area. Reviews summarise the field and usually end with a section on open problems or future work, which is a goldmine for topic ideas. As you read, keep a simple log of each paper: the question, the method, the data and the stated limitations. Gaps appear naturally when you see the same limitation mentioned repeatedly.

Step 3: Turn a Gap into a Research Question

Common ways to create an original but manageable contribution include:

• New setting: apply a well-tested method to a new population, such as Pakistani data where only Western studies exist.
• Comparison: compare several methods on the same problem under fair conditions.
• Extension: relax an assumption of an existing model, for example adding time delay to an epidemic model.
• Simulation study: test how a statistical method behaves when its assumptions are violated.
• Theory: prove a new property, bound or generalisation of a known result (more common for pure mathematics MS and PhD work).

Write your question in one sentence. If you cannot, the topic is not yet clear.

Step 4: Check Feasibility with Numbers

Many statistics and data science topics fail because the student cannot collect enough data. A quick sample size calculation, done before you commit, protects you.

Worked Example 1: Sample Size for a Survey Proportion

Suppose you want to estimate the proportion of university students in Multan who use AI tools for assignments, with a 95% confidence level and a margin of error of \( \pm 5\% \). With no prior estimate, use \( p = 0.5 \), which gives the largest (safest) sample size.

Step 1: Formula.

$$ n = \frac{z^2\, p(1 - p)}{E^2}. $$

Step 2: Substitute \( z = 1.96 \), \( p = 0.5 \) and \( E = 0.05 \).

$$ n = \frac{1.96^2 \times 0.5 \times 0.5}{0.05^2} = \frac{3.8416 \times 0.25}{0.0025} = \frac{0.9604}{0.0025} = 384.16. $$

Step 3: Round up. Always round sample sizes up, so \( n = 385 \).

Step 4: Allow for non-response. If you expect only 80% of students to respond, you need to approach \( 385/0.8 \approx 482 \), so about 482 students. If that is impossible, widen the margin of error or change the design.

Worked Example 2: Sample Size for Comparing Two Means

You plan to test whether a new teaching method improves exam scores. From earlier studies, the standard deviation of scores is about \( \sigma = 10 \) marks, and you want to detect a difference of \( \delta = 5 \) marks with a two-sided significance level of 0.05 and power of 80%.

Step 1: Formula for each group.

$$ n = \frac{2\,(z_{1-\alpha/2} + z_{1-\beta})^2\, \sigma^2}{\delta^2}. $$

Step 2: Critical values. \( z_{0.975} = 1.96 \) and \( z_{0.80} = 0.8416 \), so their sum is \( 2.8016 \) and its square is about \( 7.849 \).

Step 3: Substitute.

$$ n = \frac{2 \times 7.849 \times 100}{25} = \frac{1569.8}{25} \approx 62.8. $$

Step 4: Round up. You need 63 students per group, 126 in total. This formula uses the normal approximation; software using the t-distribution typically gives one more student per group, 64.

How to report it (APA style): "An a priori power analysis indicated that 64 participants per group would provide 80% power to detect a 5-mark difference (Cohen's \( d = 0.50 \)) at \( \alpha = .05 \), two-tailed."

Note that \( d = \delta/\sigma = 5/10 = 0.50 \), a medium effect by Cohen's conventions. You can run this in G*Power, SPSS, Stata, R or Python. In R, one line does it:

r power.t.test(delta = 5, sd = 10, sig.level = 0.05, power = 0.80)

The output gives about 63.8, which rounds up to 64 per group.

Step 5: Score and Compare Your Ideas

When you have two or three candidate topics, score each from 1 (poor) to 5 (excellent) on the key criteria. Here is an example for three ideas:

The totals are 14 for A, 15 for B and 17 for C. Topic C wins even though Topic A was the most exciting, because A depends on data that may never arrive. You may weight criteria differently, for example doubling data availability for an empirical thesis.

Step 6: Talk to Your Supervisor Early

Bring a one-page concept note: a working title, the research question, two or three key references, the proposed method, the data source and a rough timeline. Supervisors respond far better to a concrete draft than to "please suggest a topic". Their expertise and existing datasets can also make a topic far more feasible.

Promising Areas Right Now

• Statistics: causal inference, Bayesian hierarchical models, survival analysis with medical data, small-area estimation.
• Data science: explainable AI, fairness in algorithms, Urdu and regional-language NLP, forecasting electricity demand or crop yields.
• Applied mathematics: epidemic and ecological models, fractional differential equations, optimisation in logistics, numerical methods for PDEs.
• Pure mathematics: graph theory and chemical graph indices, fixed-point theory, algebraic structures, though these need close guidance from a specialist supervisor.

Common Mistakes to Avoid

• Choosing a topic only because it is fashionable.
• Ignoring data access until after the proposal is approved.
• Copying a published study without a clear new contribution.
• Planning too many objectives; two or three focused objectives are better than six vague ones.

Did You Know?

• The 384 figure from a 95% confidence, 5% margin survey calculation is so common that it appears in thousands of theses worldwide.
• Jacob Cohen's effect size conventions, small (0.2), medium (0.5) and large (0.8), were proposed as rough guides, and he warned against using them mechanically.
• Many highly cited papers began as a master's thesis that tested an established method in a new setting.

Key Takeaways

• A good topic combines interest, a genuine gap, feasibility, clear scope and practical value.
• Mine recent review articles and their "future work" sections for ideas.
• Use sample size and power calculations to test feasibility before committing.
• Score competing ideas on explicit criteria and discuss a concrete concept note with your supervisor.

If you need help choosing a topic, writing a proposal or running the analysis for your thesis, Amathify's experts and PhD scholars offer step-by-step guidance via WhatsApp or the Ask a Question form.