CHEAR: Confusion-Hierarchy Evidence-Aware Reasoning for Cost-Efficient Human Activity Recognition

Xiangqing Lu; Tao Zhu; Mingxing Nie; Liming Chen

cs.AIOct 11, 2026

CHEAR: Confusion-Hierarchy Evidence-Aware Reasoning for Cost-Efficient Human Activity Recognition

Authors: Xiangqing Lu, Tao Zhu, Mingxing Nie, Liming Chen

Abstract

Large language models (LLMs) offer a new approach to wearable-sensor-based human activity recognition. Retrieval-augmented generation (RAG) further introduces similar motion examples and discriminative sensor knowledge, enabling decisions grounded in observations relevant to the current query. More evidence, however, does not necessarily mean better-organized evidence. When online prompts and LLM agents primarily orchestrate inference, decisions about which activities to compare and when to invoke discriminative knowledge remain concentrated at inference time. As the activity vocabulary grows, the model must repeatedly process more candidates and associated knowledge, whereas pruning candidates too early risks excluding the true activity. We propose confusion-hierarchy evidence-aware reasoning (CHEAR), which moves part of this knowledge organization offline. CHEAR constructs and validates an activity hierarchy from recurrent cross-subject retrieval errors and builds corresponding activity-pair knowledge. At inference time, the hierarchy orders candidate comparisons; the LLM retains candidates progressively, and local discriminative knowledge is accessed only for the remaining activities. Across eight datasets, CHEAR achieves 86.79% mean cross-subject accuracy, a 75.91% candidate reduction rate, and 96.44% true-activity retention. Per-sample token use is 85.15% lower on average than in a variant without the hierarchical structure. Mean accuracy across eight cross-dataset transfer directions is 88.81%. These results show that organizing candidates and discriminative knowledge around observed activity confusions reduces online LLM input while maintaining recognition performance.

Figures & tables

Fig. 1: Offline knowledge construction in CHEAR.

Fig. 2: Online hierarchical inference in CHEAR.

Fig. 3: Overview of cross-subject accuracy, token cost, and results across foundation models.

	Easy						Medium						Hard				Mean
Model/Method	Opportunity		UCIHAR		Shoaib		PAMAP2		USCHAD		Mhealth		WISDM		DSADS		Mean
	Acc	F1	Acc	F1	Acc	F1	Acc	F1	Acc	F1	Acc	F1	Acc	F1	Acc	F1	Acc	F1
# Classes	4		6		7		12		12		12		18		19		–
# Channels	30		6		30		18		6		15		6		30		–
HARGPT Text	21.0	19.2	29.6	17.4	27.1	19.2	12.1	6.2	13.8	7.3	12.1	6.0	5.6	1.8	10.5	6.2	16.5	10.4
Gemini Text	26.5	19.8	24.2	13.0	27.1	17.6	15.0	10.2	14.2	5.4	25.4	20.8	11.1	7.6	13.2	8.6	19.6	12.9

Table 1: Cross-subject comparison with prior methods (accuracy and macro-F1, %). Results for prior methods are taken from ZARA [ 3 ] . Means are arithmetic means across eight datasets. For each metric in each dataset and in the mean column, the highest value is bold and the second-highest is underlined; tied values receive the same marking.

Dataset pair	Activities	Sensor-position mapping
UCIHAR $\leftrightarrow$ USCHAD	5	Waist $\leftrightarrow$ front right hip
PAMAP2 $\leftrightarrow$ WISDM	5	Wrist $\leftrightarrow$ hand
Mhealth $\leftrightarrow$ WISDM	5	Right wrist $\leftrightarrow$ hand
PAMAP2 $\leftrightarrow$ Mhealth	7	Wrist $\leftrightarrow$ right wrist
		Chest $\leftrightarrow$ chest
		Ankle $\leftrightarrow$ left ankle

Table 2: Shared activities and sensor-position mappings in cross-dataset experiments.

Transfer direction	Accuracy (%)	Macro-F1 (%)	Accuracy without transferred features (%)	Mean final candidates	Candidate reduction (%)	True-activity retention (%)
UCIHAR $\rightarrow$ USCHAD	94.00	94.00	92.00	2.270	54.60	100.00
USCHAD $\rightarrow$ UCIHAR	93.00	92.90	94.00	1.130	77.40	97.00
WISDM $\rightarrow$ PAMAP2	78.00	78.25	77.00	1.910	61.80	94.00
PAMAP2 $\rightarrow$ WISDM	92.00	91.99	88.67	1.313	73.73	98.67
WISDM $\rightarrow$ Mhealth	75.00	74.00	71.00	1.780	64.40	100.00
Mhealth $\rightarrow$ WISDM	91.33	91.38	88.67	1.400	72.00	98.00

Table 3: Cross-dataset transfer results and accuracy without transferred feature knowledge. Candidate-set metrics are aggregated over query windows.

Dataset	Full	No features	No LLM	No hierarchy
DSADS	89.47	85.79	84.74	87.37
Mhealth	88.75	87.08	86.67	88.75
Opportunity	85.00	81.00	80.00	85.00
PAMAP2	82.50	80.83	80.42	81.67
Shoaib	99.05	99.05	99.05	99.05
UCIHAR	96.67	95.00	92.08	96.25

Table 4: Cross-subject accuracy (%). Highest values are bold; ties are marked equally.

Dataset	Full	No features	No hierarchy	Savings (%)
DSADS	4,679.3	935.0	44,674.6	89.53
Mhealth	1,910.9	1,409.3	9,238.3	79.32
Opportunity	3,111.1	1,104.6	14,642.6	78.75
PAMAP2	10,038.2	1,141.6	63,655.0	84.23
Shoaib	1,013.4	787.4	8,484.7	88.06
UCIHAR	1,803.5	814.9	10,666.5	83.09

Table 5: Tokens per sample and the full method’s savings relative to the variant without the hierarchical structure.

Dataset	Root activities	Root units	Root compression (%)	Final candidates	Final reduction (%)	True-activity retention (%)
DSADS	19.00	6.00	68.42	3.09	83.71	99.47
Mhealth	12.00	8.00	33.33	3.01	74.93	100.00
Opportunity	4.00	2.00	50.00	1.55	61.38	93.00
PAMAP2	12.00	6.00	50.00	2.65	77.95	92.50
Shoaib	7.00	5.00	28.57	1.77	74.69	99.52
UCIHAR	6.00	2.00	66.67	1.80	70.00	98.33

Table 6: Hierarchical candidate reduction. Root compression is calculated from the number of root activities and root candidate units; final reduction and true-activity retention are reported together.

Dataset	Retrieval	Pair reports	Query features
DSADS	0.043	0.369	0.150
Mhealth	0.012	0.102	0.076
Opportunity	0.018	0.510	0.160
PAMAP2	0.016	0.270	0.097
Shoaib	0.019	0.244	0.106
UCIHAR	0.005	0.045	0.036

Table 7: Local processing latency of the full method (seconds per sample). Timings for individual components may overlap slightly.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Activities	Sensor positions	Activity labels
Opportunity	4	Back, upper arms, forearms	Standing, walking, sitting, lying
UCIHAR	6	Waist	Standing, sitting, lying, walking, descending stairs, ascending stairs
Shoaib	7	Left and right pockets, belt, right upper arm, right wrist	Walking, standing, jogging, sitting, cycling, descending stairs, ascending stairs
PAMAP2	12	Wrist, chest, ankle	Lying, sitting, standing, ironing, vacuum cleaning, ascending stairs, descending stairs, walking, Nordic walking, cycling, running, rope jumping
USCHAD	12	Front right hip	Sleeping, sitting, elevator down, elevator up, standing, jumping, descending stairs, walking right, walking forward, running forward, ascending stairs, walking left
Mhealth	12	Chest, right wrist, left ankle	Ascending stairs, standing still, sitting relaxed, lying, walking, bending forward, lifting arms forward, bending knees, jogging, running, jumping forward and backward, cycling

Appendix

Table A.1: Experimental datasets, sensor positions, and activity labels.

Feature family	Representative statistics or derived quantities
Time-domain statistics	Mean, standard deviation, variance, maximum, minimum, median, root mean square, peak, range, interquartile range, skewness, kurtosis, and signal magnitude area
Frequency-domain statistics	Low-, mid-, and high-frequency band powers and their ratios; dominant frequency and power; secondary peak frequency; spectral centroid, entropy, skewness, and kurtosis; weighted mean frequency; and spectral energy
Short-time Fourier transform	Maximum, mean, and standard deviation in each frequency band, plus statistics of short-time spectral entropy and spectral centroid
Change and impact	Mean, root mean square, and standard deviation of first differences; zero-crossing rate; impact rate of change; and peaks
Periodicity and dependence	First autocorrelation peak, first minimum, and first zero crossing; fourth-order autoregressive coefficients; and recurrence rate
Multiscale complexity	Wavelet-packet energy and permutation entropy

Appendix

Table D.1: Predefined features used for activity-pair knowledge.

Explore similar work

Jul 29, 2026

RAG-HAR+: Towards Cost-Efficient LLM-Based Human Activity Recognition for Edge Deployment

Human Activity Recognition (HAR) from wearable sensors supports applications in healthcare, rehabilitation, fitness tracking, and smart environments. Yet, existing deep learning approaches require dataset-specific training, large labeled corpora, and repeated adaptation to new sensor settings or activity taxonomies. Retrieval-Augmented Generation for Human Activity Recognition (RAG-HAR) addresses this by framing HAR as a training-free, retrieval-augmented task, in which statistical descriptions of sensor windows are used to retrieve similar labeled examples that guide LLM-based classification. We introduce RAG-HAR+, a retrieval-first and cost-optimized extension that strengthens retrieval while reducing dependence on LLM-based inference. RAG-HAR+ uses an offline Retrieval Designer Agent to design dataset-specific feature groups from a diverse pool of motion descriptors, enabling sensor windows to be compared using features better aligned with dataset-specific activity patterns. During inference, RAG-HAR+ uses majority voting over retrieved neighbors for samples with strong retrieval evidence and defers only uncertain cases to an LLM-based Ambiguity Resolver Agent. Across six HAR benchmarks, RAG-HAR+ maintains competitive or improved performance while reducing LLM usage, token consumption, and inference time. We further extend the RAG-HAR mobile prototype to demonstrate the practical feasibility of retrieval-first, LLM-assisted HAR in mobile sensing scenarios.

Hansi Karunarathna, Nirhoshan Sivaroopan, Chamara Madarasingha +2

University of Sydney, Australia · Curtin University, Australia · Colorado State University, USA

Human Activity Recognition Cost-Aware Inference Edge ML Retrieval-Augmented ICL

Ask PDF

Oct 11, 2026

OntoAR-HAR: Ontology-Guided Agentic RAG with Personalized Memory for Human Activity Recognition

LLMs increasingly support sensor-based human activity recognition (HAR) by linking motion signals with activity semantics and physical knowledge. Retrieval-augmented generation (RAG) grounds decisions in sensor examples and domain features, enabling training-free target-user inference with less dependence on raw numerical arrays. However, existing LLM/RAG-based HAR typically searches single-level activity spaces using static population knowledge. Consequently, they compare many unrelated candidates using the same static population evidence and overlook personalized motion patterns. We propose OntoAR-HAR, an ontology-guided agentic RAG framework combining static population and personalized knowledge. Its tree-structured activity ontology decomposes classification into local decisions among semantically related activities. The static store supplies node-constrained examples, while offline-discovered executable signal features provide numerical evidence for local distinctions. During inference, the personalized store collects selected user observations, which the LLM integrates with static knowledge and signal features while traversing the ontology to an activity label. Cross-subject and heterogeneous-sensing experiments show complementary gains from hierarchical retrieval and personalized memory across multiple LLM backbones, achieving state-of-the-art performance. By explicitly modeling hierarchical relations across activity granularities, OntoAR-HAR also addresses the long-overlooked semantic inclusion problem and avoids treating related labels as independent categories.

Xiangqing Lu, Tao Zhu, Liang Yu +6

School of Computer Science, University of South China, Hengyang, China · School of Computer Science and Technology, Dalian University of Technology, Dalian, China · Hunan University, Changsha, China +1

Human Activity Recognition Retrieval-Augmented Classification Agentic RAG Hierarchical Classification

Ask PDF

Jul 31, 2026

TRACE-TS: Attribution-Grounded and Traceable Sensor-Language Reasoning for Human Activity Understanding

Wearable sensors capture fine-grained motion patterns that support rich behavioral understanding, yet most existing methods reduce these signals to activity labels. Recent LM-based approaches generate natural-language explanations for sensor data, but their reasoning is weakly grounded in the underlying signal, leading to fluent yet unverifiable explanations. We introduce TRACE-TS (Traceable Reasoning with Attribution-Grounded Evidence), a framework for structured and signal-grounded reasoning over wearable time series. TRACE-TS uses attribution from an expert classifier to identify salient spatio-temporal sensor regions, uses them to construct DAG reasoning traces with explicit evidence provenance, and trains a compact language model to generate these traces through gated cross-attention over sensor memory tokens. At inference, the adapted model jointly outputs the activity prediction and its reasoning trace, without requiring attribution computation or teacher guidance. We introduce Semantic Node Match(SNM), an LLM-as-judge metric that diagnoses reasoning fidelity at the observation, inference, and synthesis levels, localizing hallucinated observations and broken evidence chains missed by standard NLG metrics. Across seven wearable benchmarks, TRACE-TS achieves the best average accuracy and F1 among all evaluated methods (84.43%/81.24%), and outperforms the best LLM-based baseline by 17.96% in F1. Our code is available at https://github.com/SparshRastogi/TRACE-TS.

Sparsh Rastogi, Tanmay Kumar, Baiyu Chen +3

University of New South Wales, Sydney, Australia · Thapar Institute of Engineering and Technology, Patiala, India

CoT Faithfulness Human Activity Recognition Wearable Sensing

Ask PDF