Data Science Dissertation Examples

Read this if your topic is one where the standard measure misses the thing you actually care about, a position more students are in than realize it. Kadoma's argument is that model-centric fairness metrics evaluate the system and not the person using it, so she runs studies that measure human experience instead: perception, evaluation, feelings of control and ownership while doing real work with AI tools.

The dissertation then names what those studies found that a metric could not see. Two things to copy. The framing move is stated plainly in the abstract: these systems are "sociotechnical infrastructure", not productivity aids. Every study follows from that one reframing, which is what a thesis statement is supposed to do. And the contribution is stated as both a finding and a framework, so a reader knows what to take away as well as what was discovered.

Understand, then intervene, and the title says so, which is the point. Chen studies how students make course decisions on information platforms and then builds and tests interventions on those platforms, so the dissertation moves from description to change rather than stopping at findings and recommending that somebody else act.

If you are doing applied work and have been asked "so what?", this is the structure that answers it: a first half that establishes what is happening and why, and a second that changes something and measures whether it worked. Also a useful model of studying a system you can actually modify, which is a research-design advantage worth choosing deliberately rather than stumbling into.

The mathematics entry, and the one to open if your dissertation is theorems rather than data and every example you have been shown has a methods chapter. There is no methods chapter here and that is correct. What there is instead: a short introduction, then three self-contained chapters that each open with their own contributions and, in chapters 3 and 4, their own literature review placing the work in a live research area. Each carries its results, keeps the long proofs in a section at the end of the chapter, and hands the remaining machinery to a supplement of its own at the back, pages 204 to 281.

That proportion is the thing to notice: about a quarter of the document is appendix, because a proof that would break the argument's flow belongs behind it, not in it. Read the introduction and either literature review even if the mathematics is not yours; they are a clean example of motivating technical work for a reader who does not share your specialism.

The applied-optimization entry, and the clearest example here of writing a computational method as an ordered procedure. Tan's methodology sections number the steps: train the model, use it to sample, then characterize the output, with the loss function written out and each step's inputs and outputs named. A reader could implement it.

That sounds obvious and it is rare: most computational chapters describe an approach in prose and leave the reader to reconstruct the order of operations. It also pairs theory with practice deliberately, which the title announces, so it is a usable model if your examiners want both a contribution to method and evidence that it works on a real problem.

A multi-study dissertation on whether data-driven decision-making actually serves equity in education, and a good model of holding a practical audience in view without softening the research. Williamson closes on "actionable recommendations for educators and educational organizations", and the studies are built so that those recommendations follow from evidence rather than being appended to it.

Worth reading for the shape if your work sits between a discipline and a profession: the studies answer the research question, and a separate, clearly-labelled layer translates them for practitioners. The biographical sketch is also unusually informative about how a career becomes a dissertation topic: a decade as a data architect and reporting analyst in education systems, then a doctorate examining whether that work does what it claims.