Adaptive Activity Spotting Based on Event Rates

Abstract

Abstract-To date many activity spotting approaches are static: once the system is trained and deployed it does not change anymore. There are substantial shortcomings of this approach, specifically spotting performance is hampered when patterns or sensor noise level changes.In this work an unsupervised sensitivity adaptation mechanism is proposed for activity event spotting based on expected activity event rates. The expected event rate for activity spotting was derived from the generalisation metric used in information retrieval. To illustrate generalisation effects and depict relations of spotting performance and event rate, different event rates were simulated and their precision-recall spotting performance analysed. Subsequently, the sensitivity adaptation concept is presented and evaluated. For this purpose two large datasets from personal healthcare applications were considered to explore benefits and limitations of this adaptation approach: recognition of drinking motions from inertial sensors and chewing strokes from sound. Results showed up to 28% spotting performance increase for event rate adapted operation, confirming performance benefits for sensitivity adaptation. The approach will be most applicable in situations, where estimated event rate statistics show low variance and long monitoring durations allow effective sensitivity adaptations.

Multiscale Bowel Sound Event Spotting in Highly Imbalanced Wearable Monitoring Data: Algorithm Development and Validation Study

“…In our data set, BSs were highly sparse, with BS ratios less than 0.035 across all participants, corresponding to event rates of approximately 100-300 events/hour. With event sparsity, the spotting challenge increases [ 16 ], especially when training and evaluating the spotter on different class imbalances. Despite the high class imbalance, however, EffUNet could retrieve BSs with a recall of 73% and a precision of 72%.…”

“…Retrieval generalization was additionally evaluated by analyzing PR over event rate. Event rate was defined as BS events per time unit according to Amft [ 16 ]. To compare our model performance with related work, we converted event rate to BS ratio (ie, the ratio between spectrogram time bins containing BSs and total time bins) as follows:…”

“…Nevertheless, most algorithms were tested on balanced data sets or selected subsets of the recordings only, from dozens of minutes to a few hours. However, when collecting data with a wearable device, the BS ratio of relevant events, for example, BSs versus other surrounding sounds, largely influences retrieval performance, which reflects a basic problem in pattern spotting [ 16 ]. Specifically, in naturalistic, unmodified audio data, BS events appear sparsely and their low amplitude compared with other body sounds, for example, lung sounds, hampers BS spotting.…”

RecoFit

“…To overcome the instability of HMM probabilities, [17] uses similarity to known examples as a preprocessing filter, [19] uses boosting to improve threshold models, and [1] proposes online threshold adaptation. Outside of HMMs, other proposed spotting approaches have been based on dynamic time warping [26] and string matching [32].…”