Where should you start with a data science project presentation?
First, define the research question and the main takeaway you want your audience to remember after the presentation. Then connect the data source, method, findings, and limitations of the analysis to that question. Rather than listing model names, explain which question you answered and what evidence you used. If the presentation length and evaluation criteria are unclear, confirm them in the course instructions or with the organizer.
Start preparing by gathering the problem definition, data scope, variables used, analysis outputs, and known limitations. For each finding, noting which data period and evaluation section it is based on may make it easier to check that the claim and chart are consistent in the slides. This is a suggested approach for the project; it does not, by itself, guarantee that the results are accurate.
Turn the narrative into slides
Give each slide a single purpose. The title should state not only the topic but, if possible, the slide’s main message. For example, instead of “Model results,” you could say “False positives limit the use decision,” provided that the chart genuinely supports the claim.
An adaptable example outline for a ten-minute presentation:
- Slide 1 — Title and question: Project title, presenter or team, and research question.
- Slide 2 — Problem: The context for the problem and why it is being examined.
- Slide 3 — Data: Important notes about the source, unit of observation, period, and scope.
- Slide 4 — Method: The approach and how you evaluated the results.
- Slides 5–6 — Findings: Main results and important types of errors.
- Slide 7 — Limitations: The conditions under which the result should be interpreted with care.
- Slide 8 — Conclusion: A brief answer to the research question and the next step.
This is only a planning example; adjust the number of slides and time allocation to your own instructions. The CMU course presentation guidelines suggest an introduction, data and methods, results, and conclusion/implications for a ten-minute presentation, as well as a total of 8–12 slides. This suggestion is not a required standard for every presentation (CMU course presentation guidelines). You can also read presenting data analysis results for guidance on communicating data analysis results, and science project presentations for an outline built around a scientific question, method, and findings.
Present charts as evidence
Use a chart to explain or test a specific claim. Label the axes, units, meaning of the colors, and, where necessary, the measurement period. Rework the visual for presentation size; a small screenshot taken from a notebook may be difficult to read. If uncertainty, sample size, or comparison conditions affect the main interpretation, explain them on or beside the chart.
In a classification project, a single performance metric may not describe all usage conditions. Which metrics are meaningful depends on the problem; for example, showing class-level results or types of errors can help the audience see the effects of different mistakes. Relate your choice of metrics to the project’s intended use, and do not present an outcome you did not measure as proven. The University of Minnesota page links to data science-related seminar and workshop presentations; you can review them as examples, but visuals from other projects do not validate your findings (University of Minnesota presentation archive).
Fictional example: predicting subscription cancellations
The scenario and results in this example are entirely fictional; they are not based on a real dataset or measurement. Suppose a team is presenting a project that predicts whether customers will cancel their subscriptions in the coming month. The research question could be framed as, “Can the model produce a useful list of candidates for the support team to review?” This wording does not claim that the model definitively prevents cancellations.
In the slides, first explain the decision context and data period, then the method and evaluation setup. In the results section, show, fictionally, that the model catches some cancellations but also produces false alarms in the review list. In this case, rather than summarizing the outcome with a single success label, discuss the trade-off between cases caught and the review workload created. Data scope, missing information, and the possible effects of false alarms can also be included on the limitations slide. This example therefore illustrates how a narrative can be structured, not a claim about a specific model or its performance.
Review the results and limitations
If the claim in the text does not match the chart, compare both with the analysis output. If you cannot explain the difference, revise the claim or remove it from the presentation. If you may be asked to justify your choice of method, briefly state the alternatives you actually evaluated and your selection criterion; “more advanced” is not an explanation on its own.
If it matters how the results might change across different data periods or groups, describe this as a limitation or a next step to investigate. Do not present something you have not checked as verified. If detail makes a slide crowded, preserve the main message and move the additional information to a backup slide at the end of the presentation.
Final checklist
- Is the research question clear, and is it answered on the conclusion slide?
- Is the necessary information about the data source, scope, and period provided?
- Are the charts and numbers consistent with the latest analysis output?
- Are the method and evaluation setup explained in a way the audience can understand?
- Is the metric relevant to the project question, and are important types of errors being overlooked?
- Are the limitations and the conditions under which the results apply stated?
- Are titles, axes, units, and legends readable at presentation size?
- Has the presentation been rehearsed to fit within the allotted time?
Frequently asked questions
How many slides should I prepare?
First, check the instructions and the time available. Draft an outline that covers the question, data, method, findings, and limitations, then shorten or expand it after rehearsing. The suggestion of 8–12 slides in the CMU guidelines is a recommendation for that course; it should not be treated as a rule for every event.
Should I show code in the presentation?
If the code explains an important methodological decision or analysis step, you can show a short excerpt. Otherwise, a clear summary of the data flow, method, and results may be more suitable for the main narrative. Which details are necessary depends on the audience and the purpose of the presentation.
← Back to blog
