How to Write a Research Paper as an Engineering Student

Start with what the paper has to prove

Many first-time authors treat a paper as a shorter project report. That is the most common reason drafts feel wrong. A report explains everything you built, including login screens and database tables. A paper makes one claim and gives the evidence for it. If your project is a crop disease detection app, the paper is not about the app. It is about something like whether a lightweight model trained on field photos holds up against a heavier model when images are blurred.

Before anything else, try to finish this sentence: "We show that X performs better or differently than Y, under condition Z, measured by M." If you cannot fill those blanks yet, you are not ready to draft, and that is useful to know early. The sentence will change as you learn more, but having one keeps every later decision focused.

Step 1: Narrow the topic until it can be tested

A title like "Machine Learning in Healthcare" cannot be finished in a semester. Narrow it along three dimensions: a specific problem, a specific dataset or test setup, and a specific measure of success. "Comparing two sampling methods for imbalanced diabetic retinopathy images, measured by recall on the minority class" is modest, and that is the point. You can complete it, and a reader can check it.

Check feasibility before committing. Can you get the data this week? Does your laptop or a free cloud notebook have enough compute? Can your guide review this area? A topic that fails any of these tends to collapse in the month you can least afford it.

Step 2: Read papers in layers, and keep a table

Do not read each paper from start to finish. First pass: title, abstract and conclusion, then decide whether it is relevant. Second pass: figures, tables and the method, for the papers that survive. Third pass: full detail, only for the few closest to your work. Google Scholar, IEEE Xplore and arXiv are reasonable starting points, and your college library may provide access to paid papers.

Keep one spreadsheet from day one, with a row per paper and these columns: citation, problem, dataset, method, metric and result, stated limitations, and your own note. This table becomes your related work section. It also shows you the gap, because patterns appear once the rows line up: everyone uses the same clean dataset, nobody tests on low-end hardware, or results are reported only as accuracy.

Never cite a paper from another paper's reference list without opening it. Citation errors spread that way, and an examiner may ask you about any source you list.

Step 3: Design the experiment before you run it

Write down your baseline, your metrics and your test setup before you see any results. Choosing them afterwards makes it easy to fool yourself, even unintentionally. A few checks catch the mistakes reviewers look for first:

• Baseline: compare against something established, not only against your own earlier attempt.
• Data separation: keep the test data untouched until the end, and make sure no information leaks in during preprocessing.
• Repeatability: fix random seeds, run more than once where results vary, and report the spread as well as the best run.
• Right metric: accuracy can look excellent on imbalanced data while the model misses nearly every case that matters.
• Records: note versions, parameters and dates as you go, because you will not remember them while writing.

Step 4: Draft in the order that makes writing easier

The order sections appear in is not the best order to write them. This sequence usually causes the least rewriting:

• Figures and tables first. Decide which results you will show. They are the skeleton of the paper.
• Methodology next. You already know what you did, so this is the easiest section and gets you moving.
• Results and discussion. State what happened, then what it means, including where your approach did worse.
• Related work, drawn from your spreadsheet and grouped by approach rather than listed one paper at a time.
• Introduction, once you know exactly what the paper delivers: the field, the gap, your contribution.
• Conclusion: findings, limits and realistic future work, without repeating the abstract word for word.
• Abstract and title last, because only now can you summarise the paper accurately.

Common mistakes in student drafts

Claims without evidence are the biggest problem. Replace "the proposed system is highly efficient" with what you measured and what you compared it to. Next is the padded background, where two pages explain what machine learning is. Assume your reader knows the field and spend that space on your method.

Integrity: what can and cannot be delegated

Rewriting a source by swapping words is still plagiarism if the idea or structure is not credited, and a low score on a similarity tool does not change that. A similarity report only shows text overlap. Your job is to read the flagged passages and fix real problems, not to engineer a lower number.

AI writing tools need particular care. If you are submitting to IEEE, its policy requires AI-generated content, including text, figures, images and code, to be disclosed in the acknowledgments, naming the system and the sections involved. Other publishers and many universities have their own rules, so check before you use these tools. Whatever the rule, you remain responsible for every sentence and every reference. AI tools can produce references that look real but do not exist, so verify each one against the original source.

The final revision pass

Leave a few days between finishing the draft and revising it. Then check it in rounds, one concern at a time: argument, then numbers, then language, then format. Confirm that every figure quoted in the text matches its table, that every citation appears in the reference list and the other way round, and that the title promises only what the results support.

Ask one person who was not on the project to read the abstract and introduction and tell you what they think you did. If their answer differs from your intent, the fault is in the writing. Finally, read the venue's author instructions and apply its template. If that is an IEEE venue, our IEEE paper format guide covers the details.

If you want a mentor to review your approach along the way, Road2Tech offers guidance from its base in Pune and online, with the writing and the understanding remaining yours.