Data Efficient Deep Learning Empirical Guidelines for Training Set Size Estimation in Inertial Sensor Classification

Jul 16, 2026

Deep learning models rely on large-scale inertial datasets for classification tasks such as human activity recognition and smartphone location recognition, yet collecting and labeling such data is costly, time-consuming, and difficult to scale. Existing guidance for the minimum sample size needed to achieve a target accuracy level is missing. This study empirically evaluates learning-curve convergence across inertial classification, offering a unified binary/multi-class framework, an estimation formula, and a stability-point metric. Experiments on six real-world datasets totaling 102.7 hours show accuracy grows with consistent logarithmic patterns.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

Introduction

Problem background: data requirements in HAR and SLR

Motivation: data efficiency over data volume

Knowledge gap and evaluation goal

Method overview

Unified framework for binary and multi-class settings

Empirical learning-curve analysis and estimation formula

Stability point metric for practical planning

Highlights Data-Efficient Deep Learning: Empirical Guidelines for Training Set Size Estimation in Inertial Sensor Classification Ofir Kruzel, Itzik Klein • Evaluates learning curve convergence for inertial sensor classification. • Analyzes six diverse HAR and SLR datasets totaling 102.7 hours of data. • Reveals classification accuracy consistently follows a logarithmic pattern. • Introduces a stability point metric to optimize training data collection. • Shows models reach stability with fewer samples than heuristics suggest. arXiv :2607 .09402v 1 [ cs .LG] 10 Jul 2026 Data-Efficient Deep Learning: Empirical Guidelines for Training Set Size Estimation in Inertial Sensor Classification Ofir Kruzela,∗ , Itzik Kleina a The Autonomous Navigation and Sensor Fusion Lab, the Hatter Department of Marine Technologies, University of Haifa,, Haifa, Israel ARTICLE INFO Keywords: Human Activity Recognition Smartphone Location Recognition Inertial sensors Deep learning Data efficiency Logarithmic convergence AB STRACT Deep learning models’ dependency on large-scale inertial datasets presents a significant bottleneck in inertial sensor-based classification tasks, such as human activity recognition and smartphone location recognition. In these domains, data collection requires massive recording campaigns that are complex, time-consuming, and difficult to scale. Currently, data-driven guidelines for determining the minimum sample size required to reach a desired accuracy level do not exist. To address this gap, this study presents a systematic empirical evaluation of learning curve convergence rates in inertial classification. We introduce a unified framework that analyzes classification performance under both binary and multi-class scenarios, and derive an empirical formula to estimate performance relative to dataset size. Testing across six diverse, real-world datasets totaling 102.7 hours of inertial measurements demonstrates that accuracy follows a consistent logarithmic growth pattern, regardless of task complexity. Leveraging this finding, we propose a quantitative stability point metric, defined as the sample size required for the learning curve to stabilize within a predefined mean absolute percentage deviation of its asymptotic maximum. Our analysis reveals that models often reach practical stability with substantially fewer samples than traditional heuristics suggest. Ultimately, we offer a generalizable framework to extrapolate total data requirements from small-scale pilot studies, optimizing the tradeoff between recording effort and model reliability. These findings shift the prevailing paradigm from maximizing data volume toward optimizing data efficiency, offering concrete, data-backed guidelines for planning recording campaigns in inertial sensing applications. 1. Introduction Human recognition from wearable and smartphone sensors has become a core capability in ubiquitous and mobile computing. By exploiting accelerometer and gyroscope measurements, these systems infer user activities, motion modes, and device carrying configurations CuestaVargas, Galán-Mercant and Williams (2010); Camomilla, Bergamini, Fantozzi and Vannozzi (2018), supporting applications in healthcare, sports analytics, human–computer interaction, and context aware navigation. Klein (2025); Yampolsky, Kruzel, Fekson and Klein (2025); Mardanpour, Sepahvand, Abdali-Mohammadi, Nikouei and Sarabi (2023) . Over the past decade, deep learning models, such as convolution, recurrent, and attention based architectures, have substantially improved recognition accuracy and robustness, leading to a rich body of work on model architectures, sensor configurations, and deployment scenarios Gu, Chung, Chignell, Valaee, Zhou and Liu (2021); Kaseris, Kampianakis and Tzes (2024a); Gomaa and Khamis (2023); Vertzberger and Klein (2021) . Despite these advances, comparatively little attention has been paid to a basic practical question s

FAQs

Why is training set size estimation important for inertial sensor classification?

Because collecting and labeling accelerometer/gyroscope data for tasks like HAR and SLR is labor-intensive and hard to scale, so choosing an adequate sample size is crucial to reach reliable accuracy without excessive recording effort.

What does the study find about how classification accuracy changes with dataset size?

Across six diverse real-world datasets totaling 102.7 hours, classification accuracy follows a consistent logarithmic growth pattern, largely independent of task complexity.

How does the stability point metric help with planning data collection?

It quantifies the sample size required for the learning curve to stabilize within a predefined mean absolute percentage deviation from its asymptotic maximum, allowing more efficient collection than traditional heuristics.

Jul 16, 2026

Upload by Genevieve