Analytics & ML · Study deck

Mobile Sensing: Labels and Pipeline

A phone window is only useful for activity recognition when “walking,” “stairs,” or “vehicle” refers to a clear time interval and trustworthy observation.

Data Dora is your guide for this deck.

modelingmobilesensing

Data Dora, the module guide, in a scene from this chapter.

After studying this chapter

Learning objectives

You will be able to:

Explain: The feature stage compresses that interval into a motion representation; the classifier can map only that representation to a supported class.
Explain: At 50 samples per second, a two-second window contains 50 samples/s x 2 s = 100 samples per axis.
Explain: A model trained on one pocket may fail when the phone is held or left on a desk.
Explain: Phone motion, location or speed cues, and a diary or confirmed journey can support a transport label.

Major section

Align the Label With the Motion Window

Every arrow carries a timing assumption.

If the annotator presses “walking” after movement begins, the first window may span standing and walking.
Phone motion, location or speed cues, and a diary or confirmed journey can support a transport label.
Agreement strengthens the case; disagreement should remain visible rather than being overwritten by the preferred sensor.

Activity recognition assigns one bounded label to a fixed window, so the sampling and collection context must remain attached to the result.

Major section

Align the Label With the Motion Window (continued)

It contains $(10\ \mathrm{s}\times50\ \mathrm{samples/s}=500)$ samples per axis.

Five two-second non-overlapping windows fit that segment.
Calling all five “vehicle” would attach the future state to earlier stillness.
For self-reports, account for delay and mistakes.
For scripted laboratory activities, record whether the phone position matches deployment.

A 50 Hz inertial stream becomes a bounded human-activity label through windowing, feature extraction, classification, and a confidence gate.

Major section

Start With the Story

A sampling rate is the number of sensor readings taken in one second.

A phone can take many motion readings while a person walks.
Its accelerometer senses changes in motion.
The label is a bounded guess, not a direct fact.
A model trained on one pocket may fail when the phone is held or left on a desk.

A phone fuses inertial motion with speed, road corridor, and stop spacing to distinguish bus, car, and bicycle travel.

Major section

Start With the Story (continued)

This plain path leaves out the exact feature maths and model setup.

The Practitioner section shows how to build and test the windows.
Under the Hood covers signal axes, feature values, class errors, memory, and power limits.
Include “unknown” so the model is not forced to make a false choice.
One quick tap can also be wrong.

Major section

Start With the Story (continued)

A pocket test does not prove a wrist, bag, or desk case.

A test on one group does not prove all users.
Recheck when the sensor, phone build, label set, or model changes.
Each label needs a plain meaning and a good example.
A very short window may miss the shape of a step.

Major section

Start With the Story (continued)

A long window may mix two acts and make the result late.

If an alert follows, keep that time too.
A high rate and constant model use can drain a phone or watch.
A model that works only near a charger may not fit the job.

Major section

Activity Recognition Labels

At 50 samples per second, a two-second window contains 50 samples/s x 2 s = 100 samples per axis.

The feature stage compresses that interval into a motion representation; the classifier can map only that representation to a supported class.

Deck summary

Key takeaways

Every arrow carries a timing assumption.

It contains $(10\ \mathrm{s}\times50\ \mathrm{samples/s}=500)$ samples per axis.
A sampling rate is the number of sensor readings taken in one second.
This plain path leaves out the exact feature maths and model setup.
A pocket test does not prove a wrist, bag, or desk case.

Retrieval practice

Recall check

Data Dora says: answer from memory, then check your reasoning.

Q1Why should a mobile activity recognition design record phone placement with each data collection session?

ABecause a motion window from a hand-held phone, pocket phone, backpack phone, and desk phone can represent different physical behavior.

BBecause placement can be ignored after calibrating one phone, since axes, damping, and carried motion stay equivalent across styles.

CBecause placement provides a useful proxy for the activity label, letting a pocket recording stand for walking when manual labels are sparse.

DBecause matching sample rate and window length makes walking records comparable across phones without a separate placement check.

Show answer

Answer: A Phone placement is part of the evidence boundary for inertial activity recognition.

Print reference

Answers

Answer key.

A · Phone placement is part of the evidence boundary for inertial activity recognition.