False positive rate (FPR)
GM of Operations
False positive rate (FPR) is the proportion of actual negative cases incorrectly flagged as positive. Calculate it as false positives divided by false positives plus true negatives. For camera reviews, the percentage of flagged alerts found to be false uses a different denominator.
A false positive is a detection raised when the condition being tested is absent. A true negative is a negative case correctly left unflagged. The false positive rate measures errors across those actual negative cases: FPR = FP / (FP + TN). Multiply by 100 to express it as a percentage. The test must define what counts as one case, such as a labelled driving interval, and how the reference outcome was established.
Three metrics, two different denominators
FPR asks how often negative cases are wrongly flagged. Precision asks how many flagged cases are correct. The share of flagged cases found false is the false-discovery proportion, or one minus precision.
| Metric | Calculation | What is counted below the line |
|---|---|---|
| False positive rate | FP / (FP + TN) | All actual negative cases |
| Precision | TP / (TP + FP) | All flagged cases |
| Share of alerts found false | FP / (TP + FP) | All flagged cases |
TP means true positive, FP means false positive and TN means true negative. A zero denominator makes a metric undefined; it isn't evidence of a zero error rate.
Sources: Google Machine Learning Crash Course: classification metrics · Scikit-learn: precision, recall and classification metrics
A worked example, not a fleet benchmark
Imagine a reference-labelled test with 1,000 cases where the target condition is absent. The system correctly leaves 980 unflagged and raises 20 false alerts. It also correctly flags 20 cases where the condition is present.
Its false positive rate is 20 / (20 + 980) = 2%. Across the 40 alerts it raises, 20 are false, so the share of alerts found false is 50%. Precision is also 50%. The same example produces different percentages because the denominators answer different questions.
These are illustrative counts, not observations from MapTrack or a camera study. If you've only reviewed a selected queue, describe that sample and avoid presenting its result as the rate across all driving.
A valid alert can still need no follow-up
Separate the detected condition from the review decision. An event can meet a defined trigger while needing no action under your review policy. That differs from an incorrect detection. Keep unavailable or inconclusive footage in a separate group.
Sources: FMCSA: Driver Distraction in Commercial Vehicle Operations, section 4.1.2
Why it matters
An alert queue shows what was flagged. It doesn't show every negative case the system correctly left alone. Before comparing results, ask which events, time windows and settings were tested, who checked the reference labels, and how missing footage was handled. Keep each alert type separate. A useful fleet review also records how many alerts need follow-up and how long that work takes. A detection metric alone doesn't measure that workload.
How MapTrack helps
With the connected-camera setup enabled, authorised reviewers can view live video and saved 30-second events in app.maptrack.com alongside the asset record, pre-starts and maintenance. Confirm configuration and reviewer access before rollout, then use a defined assessment process when evaluating alerts.
Related guides
Know where every asset is, right now
QR, GPS and barcode tracking in one platform. See real-time location and status.
- No credit card required
- 30 days free trial
- Cancel anytime
Frequently asked questions
Can I calculate false positive rate from a camera alert queue?
Not from the queue alone. You need the actual negative cases, including those correctly left unflagged. A reviewed alert queue can tell you the share of assessable alerts found false in that sample, provided you state how the sample was selected.
What is a good false positive rate for a fleet camera?
There isn't a universal pass mark. Assess the alert type, reference labels, test conditions, missed detections and review workload together. Record the settings and sample size so you're comparing like with like.
Does an alert needing no action count as a false positive?
Only if the detection is wrong under the stated definition. An accurately detected event may need no follow-up under your policy. Record detection validity and the follow-up decision separately.
Related terms
AI dash cam
An AI dash cam is a connected vehicle camera with enough on-device processing to run computer-vision models against the video stream in real time. Road-facing models typically cover following distance, lane departure and forward collision risk. Driver-facing models cover distraction, phone use, drowsiness, smoking and seatbelt use. The device decides which moments are worth keeping, saves a short clip either side of the trigger, and uploads it with the matching GPS and sensor data. The word AI describes where the analysis happens, not a difference in image quality.
Driver monitoring system (DMS)
A driver monitoring system points inward. Using an infrared cabin-facing camera so it works at night and through sunglasses, it tracks head pose, eye state and posture to infer whether the driver is attentive. It is the counterpart to ADAS, which points outward at the road. In fleet equipment the two are usually sold together in a single dual-facing unit, and the terms are often used loosely as though they were one feature.
Video telematics
Video telematics combines road-facing and sometimes driver-facing camera footage with the telematics data recorded at the same moment: GPS position, speed, heading, acceleration and time. Traditional telematics tells you that a harsh braking event occurred at a location and a speed. Video telematics shows you what was in front of the vehicle when it happened. Most systems record continuously to local storage and upload only short clips around a triggered event, because uploading every minute of footage over a mobile network is neither affordable nor useful.
Cite this definition
Writing about this topic? You’re welcome to quote this definition. Here’s the wording to use so your readers can find the original.
A false positive is a detection raised when the condition being tested is absent.
- Short attribution
- MapTrack Glossary: False positive rate (FPR), https://www.maptrack.com/glossary/false-positive-rate
- Reference list
- MapTrack. (2026). False positive rate (FPR) [Glossary definition]. Retrieved from https://www.maptrack.com/glossary/false-positive-rate
Free to reuse with credit under a Creative Commons Attribution 4.0 licence. If you’d rather link straight to it, the page is https://www.maptrack.com/glossary/false-positive-rate.
See how MapTrack handles false positive rate (fpr)
Ready to track every asset?
Join construction, mining and field service teams across Australia.
4.9 on G2 · 5.0 on Capterra
- No credit card required
- 30 days free trial
- Cancel anytime