How to Present Data Visualisations in a University Assignment

A heatmap where every cell is the same shade. A confusion matrix pasted in with no explanation of the classes. Markers on data science modules see these every term, and they suggest the analysis was not understood. Data visualisation in a data science assignment means using plots to explore your dataset, justify your modelling choices and prove your results. This guide shows UK students which plots to use at each stage, how to format them, and how to explain them for 2:1 and first-class marks.

What Markers Expect From Data Visualisations in a Data Science Assignment

Plots serve different purposes at different stages of a project. Exploratory plots show you understand the dataset, including its shape, gaps and outliers. Diagnostic plots justify choices like scaling a feature or dropping a column, Results plots prove the model actually works.

However, what UK marking rubrics reward is critical evaluation and reproducibility. A plot with no link to a decision earns little credit. A 2:1 submission shows a correlation heatmap. A first uses it to explain why two features were merged to reduce multicollinearity.

You have to be selective about what stays in the main body. Keep the plots that support your argument, and move extras, such as a full pair plot, to an appendix, with a clear reference in the text. Check your module handbook too, since figures often sit outside the word count while captions and written discussion count.

Knowing the purpose of each plot makes it much easier to pick the right one.

Match Each Plot to the Stage of Your Analysis

Here are the plots you should use to answer the question your analysis is asking at that point.

• Histogram or box plot → Checks distribution and outliers before cleaning, such as skew in HM Land Registry house price data.
• Time series line plot → Shows trend and seasonality, such as monthly retail sales from the Office for National Statistics.
• Correlation heatmap → Flags linked features, which supports your feature selection decisions.
• Scatter plot with trend line → Tests a relationship before regression, such as study hours against exam scores.
• Residual plot → Shows whether regression errors look random, which checks your model assumptions.
• Confusion matrix → Shows how often each class is wrongly predicted in a classification model.
• ROC, or precision-recall curve → Compares classifiers across thresholds, and precision-recall suits imbalanced data better.
• Learning curve → Reveals overfitting by comparing training and validation scores as data grows.
• Feature importance chart → Explains which inputs drive predictions, which supports interpretability.

Avoid pie charts and 3D plots, because they hide comparisons. Once you have the right plots, formatting decides whether markers trust them.

Format Plots So They Are Readable and Reproducible

Clear formatting protects easy marks, and reproducible code shows technical honesty.

• Caption and number: Label items as Figure 1 with a caption stating what the plot shows and which data split it uses. Many universities place figure captions below, so check your style guide.
• Axes and units: Label both axes and state units, such as “Median house price ( £)”.
• Colour: Use colour-blind safe palettes such as viridis, since about 1 in 12 men has colour vision deficiency.
• Accessibility: Vary line styles or marker shapes so meaning never depends on colour alone, and add alt text for digital submissions.
• Scale: Start bar chart axes at zero, and use a log scale only when your caption says so.
• Data ethics: Anonymise personal data to meet UK GDPR, and never crop an axis to exaggerate a result.
• Export: Save plots at 300 dpi or as SVG, then size them so labels match your body text.
• Reproducibility: Set a random seed, keep plotting code in an appendix or repository, and cite data sources in Harvard style, for example (ONS, 2025).

Python users can build most of these plots with matplotlib and seaborn, and R users with ggplot2. Tableau and Power BI suit dashboards, but they make reproducibility harder, so record your export settings.

Explain Every Plot With Evidence

Your writing has to turn each plot into an argument. So mention the figure before it appears, state the key finding with a number and then say that decision it led to. For example” “Figure 4 shows recall for the minority class at 0.58, so I rebalanced the training data using SMOTE." That single sentence shows evidence and judgement together.

When you compare models, show the results side by side in one table or one chart. The marker should see which model won without hunting through pages. After that, add one limit. Perhaps the test set is small, or the ROC curve hides weak precision on imbalanced data. Naming a limit is what moves work from a 2:1 towards a first.

Run the final checks below before you submit.

Pre-Submission Checks

Go through this list on your final draft, one plot at a time.

• Each plot links to a decision or finding in the text.
• Every figure is mentioned before it appears.
• Axes, units, caption and data source are all present.
• Plots stay readable in greyscale and at printed size.
• Numbers in your text match the plot exactly.
• Your code runs from start to finish and reproduces every figure.

Nowadays students also seek data science assignment support before final submission. You can also prefer to review your plots from a subject professional to avoid mark deduction.