Curious to delve into the statistics of your newly created classification? Follow these steps:
Select your classification in the table of contents.
From the top menu bar, click on the
Calculate statisticsbutton.
A pop up will appear with a range of statistics for you to explore and interact with, depending on the type of classification that you performed (using training samples or thresholds).
Alternatively, you can access the calculate statistics tool via the toolbox > analysis > calculate statistics.
Understanding your data
Click on the expand arrow at the top right any graph to expand it - this will make it easier to see and interact with.
You can show or hide the results for each class on the graphs by clicking its name in the legend. A class will be 'greyed out' in the legend when you have hidden it.
Total area
The Donut Chart visually compares the area coverage of each category in the classification. The total area of all classes is displayed in the center. Hovering over a segment of the graph reveals the class name and its area. This helps you quickly understand how different classes contribute to the overall area distribution.
Overall accuracy
* Not available for threshold classifications
Overall accuracy is a measure of how well your classification performed, shown as a percentage. It’s calculated by comparing all your sample points—both training and validation—to the results of the classification. We check how many points were correctly classified and divide that by the total number of points.
For example, if you created 100 sample points and 87 of them were correctly matched to the right class in the final classification, your overall accuracy would be 87%.
In GeoNadir, we automatically split your input sample points: 70% are used to train the model, and 30% are used to test it—but the final accuracy is based on all 100% of your points to give you a clear picture of how well the model performed overall.
Sankey chart - accuracy
* Not available for threshold classifications
The Sankey chart helps you see how well each class in your dataset was classified.
Each colored bar on the left shows the true class of your sample points, while the bars on the right show how those points were predicted by the classification model. The lines (or “flows”) between them show where the points went—correct matches flow straight across, and misclassifications flow into the wrong class.
The thicker the line, the more points it represents. You can hover over each bar or line to see the class name and number of points.
This is a great way to spot which classes were most accurate and where confusion occurred between similar-looking areas.
Compare
* Not available for threshold classifications
The Lollipop chart gives you a quick visual comparison of classification results per class. You can switch between different metrics to explore how well each class performed:
Class accuracy – the percentage of points in each class that were correctly classified.
False positives – points incorrectly predicted as this class.
False negatives – points that belong to this class but were misclassified as something else.
Total area – the total area that the model assigned to this class. The statistics unit for the area measurements (e.g. m2, ft2) will be the same as your project units. If you would like to change your measurement units, follow these instructions.
This makes it easy to spot which classes performed well and where the model may have over- or under-predicted.
Below the chart, you'll find a summary table showing the same statistics in detail, so you can dig deeper into the numbers if needed. Use the expand arrow to the top right of the graph to expand to a larger version.
To change the statistic displayed on the large graph, simply click on the column in the table. The highlighted column in the table will match the y-axis of the graph.
