Training-free models struggle to recognize human activities accurately
More Features Are Not More Evidence: Limits of Training-Free Human Activity Recognition with Jev
Artificial IntelligenceMachine Learning
Summary
Identifying what people are doing from sensor data usually requires training models with examples. The authors tested a model called Jev that tries to classify activities without any training, relying on fixed rules instead. They found that Jev performs much worse than models that learn from labeled data. Adding more features to the sensor data did not help Jev and sometimes made it worse. This shows that how sensor data is presented to a model is very important, especially for models without training.
What this means in practice
- •For wearable device developers: Assess the limitations of training-free models when building activity recognition features without labeled data.
- •For mobile health app teams: Evaluate sensor preprocessing methods to improve activity classification accuracy with minimal training.
Authors
Orhan Konak
Abstract
General-purpose models promise sensor-based decisions without training a task-specific classifier, which could reduce the dependence of Human Activity Recognition (HAR) on labeled data. Yet it remains unclear whether such models can directly interpret deterministic descriptions of physical sensor signals well enough to replace or complement trained HAR models. We study this question using Jev, a fixed general-purpose probabilistic decision model, on 1,800 class-balanced accelerometer windows from WISDM, UCI341, and PAMAP2. Jev receives no labeled examples, retrieval context, or HAR-specific parameter updates. We evaluate three deterministic sensor representations and compare 5,400 Jev decisions with a generative baseline and three supervised HAR models. Jev remains far below supervised recognition, with its strongest representation reaching macro-F1 of 0.038, 0.118, and 0.089 across the three datasets, compared with 0.686 to 0.907 for the supervised models. More numerical features do not improve Jev. Instead, they reduce recognition on all three datasets, while augmenting the same numerical evidence with a deterministic semantic rendering partially recovers performance, although the experiment does not isolate semantics from the accompanying serialization and redundancy changes. Jev is fast and inexpensive to query, but its probabilities are not reliably calibrated for recognition. A post-hoc fusion analysis finds a small improvement on WISDM that does not replicate on UCI341 or PAMAP2. These results show that training-free sensor decisions depend not only on the information available in the signal, but also on whether the model can use the representation through which that information is exposed. The sensor-to-model interface should therefore be treated as part of the model evaluation rather than as a neutral preprocessing step.