11.5 Sensor and IoT time-series
You can model data coming off real hardware, with its real defects.
Before:06. Deep Learning
Sensor and IoT streams are time series with hardware defects included: drift, dropouts, desynchronised clocks, calibration error. Windowing and labelling strategies decide what a model can learn, with activity recognition as the template problem. It sits in the bridge module because hardware-adjacent data is ECE home ground. The leakage trap is mechanical: window before splitting and overlapping windows straddle the boundary, sharing samples between train and test invisibly.
Work through these
Multivariate sensor streams and synchronization
Several sensors producing readings at different rates, which must be aligned in time before anything else. Synchronisation is the unglamorous step that most projects underestimate.
Segmentation, windowing and labelling
Cutting a continuous stream into windows and attaching labels to them, which is where most of the modelling decisions actually sit. Window length frequently matters more than the model.
Human activity recognition as a template problem
Recognising what a person is doing from wearable sensors, which is the standard worked example because it has every characteristic problem in it. Solving it once transfers widely.
Sensor drift, dropouts and calibration
Real sensors drift out of calibration, drop readings, and fail quietly. Handling these is not a preprocessing detail; it is most of the engineering.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.