Smartphone Inertial Sensors Research Articles

A benchmark for domain adaptation and generalization in smartphone-based human activity recognition

Human activity recognition (HAR) using smartphone inertial sensors, like accelerometers and gyroscopes, enhances smartphones’ adaptability and user experience. Data distribution from these sensors is affected by several factors including sensor hardware, software, device placement, user demographics, terrain, and more. Most datasets focus on providing variability in user and (sometimes) device placement, limiting domain adaptation and generalization studies. Consequently, models trained on one dataset often perform poorly on others. Despite many publicly available HAR datasets, cross-dataset generalization remains challenging due to data format incompatibilities, such as differences in measurement units, sampling rates, and label encoding. Hence, we introduce the DAGHAR benchmark, a curated collection of datasets for domain adaptation and generalization studies in smartphone-based HAR. We standardized six datasets in terms of accelerometer units, sampling rate, gravity component, activity labels, user partitioning, and time window size, removing trivial biases while preserving intrinsic differences. This enables controlled evaluation of model generalization capabilities. Additionally, we provide baseline performance metrics from state-of-the-art machine learning models, crucial for comprehensive evaluations of generalization in HAR tasks.

Detection of aberration in human behavior using shallow neural network over smartphone inertial sensors data

AbstractThe integration of different Mobile Edge Computing (MEC) applications has significantly enhanced the realm of security and surveillance, with Human Activity Recognition (HAR) standing out as a crucial application. The diverse sensors found in smartphones have made it convenient for monitoring applications to gather and analyze data, rendering them valuable for HAR purposes. Moreover, MEC can be employed to automate surveillance, allowing intelligent monitoring of restricted areas to identify and respond to unwanted or suspicious activities. This research develops a system using motion sensors in smartphones to identify unusual human activities. People's smartphones were employed to monitor both suspicious and regular activities. Information was collected for various actions categorized as either suspicious or regular. When a person performs a certain action, the smartphone records a series of sensory data, analyse important patterns from the basic data, and then determines what the person is doing by combining information from different sensors. To prepare the data, information from different sensors was aligned to a shared timeline. In this study, we used a sliding window approach on synchronized data to feed sequences into LSTM and CNN models. These models, which include initial layers of LSTM and CNN, automatically find important patterns in the order of human activities. We combined SVM with the features extracted by the shallow Neural Network to make a mixed model that predicts suspicious activities. Lastly, we compared LSTM, CNN, and our new shallow mixed neural network using a new real‐time dataset. The mixed model of CNN and SVM achieved an accuracy of 94.43%. Additionally, the sliding window method's effectiveness was confirmed with a 4.28% improvement in accuracy.

L Test Subtask Segmentation for Lower-Limb Amputees Using a Random Forest Algorithm.

Functional mobility tests, such as the L test of functional mobility, are recommended to provide clinicians with information regarding the mobility progress of lower-limb amputees. Smartphone inertial sensors have been used to perform subtask segmentation on functional mobility tests, providing further clinically useful measures such as fall risk. However, L test subtask segmentation rule-based algorithms developed for able-bodied individuals have not produced sufficiently acceptable results when tested with lower-limb amputee data. In this paper, a random forest machine learning model was trained to segment subtasks of the L test for application to lower-limb amputees. The model was trained with 105 trials completed by able-bodied participants and 25 trials completed by lower-limb amputee participants and tested using a leave-one-out method with lower-limb amputees. This algorithm successfully classified subtasks within a one-foot strike for most lower-limb amputee participants. The algorithm produced acceptable results to enhance clinician understanding of a person's mobility status (>85% accuracy, >75% sensitivity, >95% specificity).

Mobile App for Enhanced Anterior Cruciate Ligament (ACL) Assessment in Conscious Subjects: "Pivot-Shift Meter".

Anterior cruciate ligament (ACL) instability poses a considerable challenge in traumatology and orthopedic medicine, demanding precise diagnostics for optimal treatment. The pivot-shift test, a pivotal assessment tool, relies on subjective interpretation, emphasizing the need for supplementary imaging. This study addresses this limitation by introducing a machine learning classification algorithm integrated into a mobile application, leveraging smartphones' built-in inertial sensors for dynamic rotational stability assessment during knee examinations. Orthopedic specialists conducted knee evaluations on a cohort of 52 subjects, yielding valuable insights. Quantitative analyses, employing the Intraclass Correlation Coefficient (ICC), demonstrated robust agreement in both intraobserver and interobserver assessments. Specifically, ICC values of 0.94 reflected strong concordance in the timing between maneuvers, while signal amplitude exhibited consistency, with the ICC ranging from 0.71 to 0.66. The introduced machine learning algorithms proved effective, accurately classifying 90% of cases exhibiting joint hypermobility. These quantifiable results underscore the algorithm's reliability in assessing knee stability. This study emphasizes the practicality and effectiveness of implementing machine learning algorithms within a mobile application, showcasing its potential as a valuable tool for categorizing signals captured by smartphone inertial sensors during the pivot-shift test.

Intra and Inter-Device Reliabilities of the Instrumented Timed-Up and Go Test Using Smartphones in Young Adult Population.

The Timed-Up and Go (TUG) test is widely utilized by healthcare professionals for assessing fall risk and mobility due to its practicality. Currently, test results are based solely on execution time, but integrating technological devices into the test can provide additional information to enhance result accuracy. This study aimed to assess the reliability of smartphone-based instrumented TUG (iTUG) parameters. We conducted evaluations of intra- and inter-device reliabilities, hypothesizing that iTUG parameters would be replicable across all experiments. A total of 30 individuals participated in Experiment A to assess intra-device reliability, while Experiment B involved 15 individuals to evaluate inter-device reliability. The smartphone was securely attached to participants' bodies at the lumbar spine level between the L3 and L5 vertebrae. In Experiment A, subjects performed the TUG test three times using the same device, with a 5 min interval between each trial. Experiment B required participants to perform three trials using different devices, with the same time interval between trials. Comparing stopwatch and smartphone measurements in Experiment A, no significant differences in test duration were found between the two devices. A perfect correlation and Bland-Altman analysis indicated good agreement between devices. Intra-device reliability analysis in Experiment A revealed significant reliability in nine out of eleven variables, with four variables showing excellent reliability and five showing moderate to high reliability. In Experiment B, inter-device reliability was observed among different smartphone devices, with nine out of eleven variables demonstrating significant reliability. Notable differences were found in angular velocity peak at the first and second turns between specific devices, emphasizing the importance of considering device variations in inertial measurements. Hence, smartphone inertial sensors present a valid, applicable, and feasible alternative for TUG assessment.

Human Activity Recognition Using Inertial Sensors in a Smartphone: Technical Background(Review)

Human Activity Recognition (HAR) stands at the intersection of machine learning, deep learning, and sensor technology, primarily focusing on leveraging inertial sensors in smartphones and wearable devices. This paper presents a comprehensive technical overview of HAR, examining the amalgamation of machine learning and deep learning systems while considering the data inputs from mobile and wearable inertial sensors. The review encompasses a broad spectrum of methodologies applied to HAR, ranging from classical machine learning algorithms to cutting-edge deep learning architectures. Emphasis is placed on the nuanced challenges and opportunities posed using inertial sensors in smartphones and wearables. This includes discussions on data preprocessing strategies, feature extraction methods, and model architectures, accounting for the unique characteristics of sensor data, such as noise, variability, and power consumption. The paper explores recent advancements, scrutinizing state-of-the-art approaches, innovative model architectures, and emerging trends in HAR. Through a comparative evaluation of various machine learning and deep learning techniques, the review aims to guide researchers and practitioners in selecting the most appropriate methods for HAR applications across diverse scenarios. In conclusion, this paper serves as an inclusive guide to the technical landscape of HAR, incorporating insights from both mobile and wearable inertial sensors. By synthesizing existing knowledge and addressing future research directions, it aims to propel advancements in developing robust and efficient systems for recognizing human activities, accommodating the evolving landscape of sensor technologies in mobile and wearable devices.

Human Activity Recognition through Smartphone Inertial Sensors with ML Approach

Human Activity Recognition (HAR) has several applications in healthcare, security, and assisted living systems used in smart homes. The main aim of these applications or systems is to classify body movement read from the built in sensors such as accelerometers and gyroscopes. Some actions could be performed in response to the output of these HAR systems. The number of smartphone users increases, whereas the sensors are widely available in different sizes and shapes (internal or external sensors). Recent advances in sensor technology and machine learning have led researchers to conduct studies on sensor technology such as HAR. HAR systems typically use a combination of sensors, such as accelerometers, gyroscopes, and cameras, to collect images or signal data that can be classified by machine learning algorithms. HAR research has focused on several key challenges including dealing with variability in sensor data, handling missing data or noise, and dealing with large amounts of sensor-generated data. In this work, several machine learning algorithms were tested in predefined settings using the KU-HAR dataset in a series of experiments. Subsequently, various performance metrics were calculated to assess the chosen algorithms’ performance. The experimental findings showed that the LightGBM classifier surpassed the other machine learning algorithms in performance metrics, such as accuracy, F1 score, precision, and recall. Although Gradient Boosting has lengthy training time, the other classifiers complete their training in an acceptable time period.

Driving Behaviour Estimation System Considering the Effect of Road Geometry by Means of Deep NN and Hotelling Transform

In this work, an intelligent hybrid model is proposed to identify hazardous or inattentive driving manoeuvres on roads, with the final goal being to increase and ensure travellers’ safety and comfort. The estimation is based on the effects that road geometry may have on vehicle accelerations, displacements and dynamics. The outputs of the intelligent systems proposed are how the type of driving can be characterized as normal, careless or distracted. The intelligent system consists of an LSTM (Long Short-Term Memory) neural network in a first step that distinguishes between normal and abnormal driving behaviour and then a second module that classifies abnormal forms of driving as aggressive or inattentive, with the latter implemented with another LSTM, a CNN (convolutional neural network) or the Hotelling transform. They are applied to some of the characteristics of vehicle dynamics to estimate the driving behaviour. Smartphone inertial sensors such as GPS, accelerometers and gyroscopes are used to measure these vehicle characteristics and to identify driving events in manoeuvres. Specifically, the critical acceleration due to the influence of the road geometry can be measured with inertial sensors, and then, this road acceleration with the lateral acceleration allows us to estimate the driver’s perceived acceleration. This perceived acceleration affects the driving style and, consequently, the estimation of the appropriate speed to travel on that road. There is use of both a traditional two-lane and a motorway route located in the Madrid region of Spain. Driving behaviour is determined by considering how changes in road geometry may affect one’s driving style and, consequently, the estimation of the proper speed. The results obtained with some of the proposed configurations of the intelligent hybrid system reach an accuracy of 97.21% in detecting dangerous driving or driving with a certain risk. This could allow generating real-time alerts for potentially dangerous or inattentive manoeuvres, leading to safer and more appropriate driving.

Smartphone-based evaluation of static balance and mobility in long-lasting COVID-19 patients.

SARS-CoV-2 infection can lead to a variety of persistent sequelae, collectively known as long COVID-19. Deficits in postural balance have been reported in patients several months after COVID-19 infection. The purpose of this study was to evaluate the static balance and balance of individuals with long COVID-19 using inertial sensors in smartphones. A total of 73 participants were included in this study, of which 41 had long COVID-19 and 32 served as controls. All participants in the long COVID-19 group reported physical complaints for at least 7 months after SARS-CoV-2 infection. Participants were evaluated using a built-in inertial sensor of a smartphone attached to the low back, which recorded inertial signals during a static balance and mobility task (timed up and go test). The parameters of static balance and mobility obtained from both groups were compared. The groups were matched for age and BMI. Of the 41 participants in the long COVID-19 group, 22 reported balance impairment and 33 had impaired balance in the Sharpened Romberg test. Static balance assessment revealed that the long COVID-19 group had greater postural instability with both eyes open and closed than the control group. In the TUG test, the long COVID-19 group showed greater acceleration during the sit-to-stand transition compared to the control group. The smartphone was feasible to identify losses in the balance motor control and mobility of patients with long-lasting symptomatic COVID-19 even after several months or years. Attention to the balance impairment experienced by these patients could help prevent falls and improve their quality of life, and the use of the smartphone can expand this monitoring for a broader population.

Meta-heuristic based feature selection for aberration detection in human activity using smartphone inertial sensors

Meta-heuristic based feature selection for aberration detection in human activity using smartphone inertial sensors

Smartphone Technology to Remotely Measure Postural Sway during Double- and Single-Leg Squats in Adults with Femoroacetabular Impingement and Those with No Hip Pain.

The COVID-19 pandemic has accelerated the demand for utilising telehealth as a major mode of healthcare delivery, with increasing interest in the use of tele-platforms for remote patient assessment. In this context, the use of smartphone technology to measure squat performance in people with and without femoroacetabular impingement (FAI) syndrome has not been reported yet. We developed a novel smartphone application, the TelePhysio app, which allows the clinician to remotely connect to the patient's device and measure their squat performance in real time using the smartphone inertial sensors. The aim of this study was to investigate the association and test-retest reliability of the TelePhysio app in measuring postural sway performance during a double-leg (DLS) and single-leg (SLS) squat task. In addition, the study investigated the ability of TelePhysio to detect differences in DLS and SLS performance between people with FAI and without hip pain. A total of 30 healthy (nfemales = 12) young adults and 10 adults (nfemales = 2) with diagnosed FAI syndrome participated in the study. Healthy participants performed DLS and SLS on force plates in our laboratory, and remotely in their homes using the TelePhysio smartphone application. Sway measurements were compared using the centre of pressure (CoP) and smartphone inertial sensor data. A total of 10 participants with FAI (nfemales = 2) performed the squat assessments remotely. Four sway measurements in each axis (x, y, and z) were computed from the TelePhysio inertial sensors: (1) average acceleration magnitude from the mean (aam), (2) root-mean-square acceleration (rms), (3) range acceleration (r), and (4) approximate entropy (apen), with lower values indicating that the movement is more regular, repetitive, and predictable. Differences in TelePhysio squat sway data were compared between DLS and SLS, and between healthy and FAI adults, using analysis of variance with significance set at 0.05. The TelePhysio aam measurements on the x- and y-axes had significant large correlations with the CoP measurements (r = 0.56 and r = 0.71, respectively). The TelePhysio aam measurements demonstrated moderate to substantial between-session reliability values of 0.73 (95% CI 0.62-0.81), 0.85 (95% CI 0.79-0.91), and 0.73 (95% CI 0.62-0.82) for aamx, aamy, and aamz, respectively. The DLS of the FAI participants showed significantly lower aam and apen values in the medio-lateral direction compared to the healthy DLS, healthy SLS, and FAI SLS groups (aam = 0.13, 0.19, 0.29, and 0.29, respectively; and apen = 0.33, 0.45, 0.52, and 0.48, respectively). In the anterior-posterior direction, healthy DLS showed significantly greater aam values compared to the healthy SLS, FAI DLS, and FAI SLS groups (1.26, 0.61, 0.68, and 0.35, respectively). The TelePhysio app is a valid and reliable method of measuring postural control during DLS and SLS tasks. The application is capable of distinguishing performance levels between DLS and SLS tasks, and between healthy and FAI young adults. The DLS task is sufficient to distinguish the level of performance between healthy and FAI adults. This study validates the use of smartphone technology as a tele-assessment clinical tool for remote squat assessment.

Estimating the Standing Long Jump Length from Smartphone Inertial Sensors through Machine Learning Algorithms.

The length of the standing long jump (SLJ) is widely recognized as an indicator of developmental motor competence or sports conditional performance. This work aims at defining a methodology to allow athletes/coaches to easily measure it using the inertial measurement units embedded on a smartphone. A sample group of 114 trained young participants was recruited and asked to perform the instrumented SLJ task. A set of features was identified based on biomechanical knowledge, then Lasso regression allowed the identification of a subset of predictors of the SLJ length that was used as input of different optimized machine learning architectures. Results obtained from the use of the proposed configuration allow an estimate of the SLJ length with a Gaussian Process Regression model with a RMSE of 0.122 m in the test phase, Kendall's τ < 0.1. The proposed models give homoscedastic results, meaning that the error of the models does not depend on the estimated quantity. This study proved the feasibility of using low-cost smartphone sensors to provide an automatic and objective estimate of SLJ performance in ecological settings.

An efficient deep learning-based approach for human activity recognition using smartphone inertial sensors

Human activity recognition (HAR) has recently witnessed outstanding growth in health and entertainment applications. Owing to the availability of smartphones, many new methods and protocols for using the data from smartphones’ embedded sensors are emerging. Nonetheless, the methods carried out and published in the literature leave a wide area for improvement, in terms of accuracy, resource economy, and adaptation to real-world nuisances. On top of that, a novel classification method that is more economical and efficient is proposed in this paper using both 1D convolutional neural network (1D-CNN) parameters and handcrafted temporal and frequency features with the proficiency of a multilayer perceptron neural network (MLP) classifier. The method proposed requires only tri-axial accelerometer data, allowing it to be deployed even into lower equipment devices; it was tested within the two well-known benchmark datasets: UCI-HAR and Uni-MIB SHAR. Experimental results yield a classification accuracy exceeding 99%, outperforming many of the methods recently shown in the literature.

RadarPDR: Radar-Assisted Indoor Pedestrian Dead Reckoning

Pedestrian dead reckoning (PDR) is the critical component in indoor pedestrian tracking and navigation services. While most of the recent PDR solutions exploit in-built inertial sensors in smartphones for next step estimation, due to measurement errors and sensing drift, the accuracy of walking direction, step detection, and step length estimation cannot be guaranteed, leading to large accumulative tracking errors. In this paper, we propose a radar-assisted PDR scheme, called RadarPDR, which integrates a frequency-modulation continuous-wave (FMCW) radar to assist the inertial sensors-based PDR. We first establish a segmented wall distance calibration model to deal with the radar ranging noise caused by irregular indoor building layouts and fuse wall distance estimation with acceleration and azimuth signals measured by the inertial sensors of a smartphone. We also propose a hierarchical particle filter(PF) together with an extended Kalman filter for position and trajectory adjustment. Experiments have been conducted in practical indoor scenarios. Results demonstrate that the proposed RadarPDR is efficient and stable and outperforms the widely used inertial sensors-based PDR scheme.

Online Learning And Saliency Effects On CNN-Based Gait Recognizers

Authentication through gait analysis offers a reliable and aneasy-to-use alternative to common authentication methods. This paper presents a novel gait recognizer that exploits online learning in Convolutional Neural Network, CNN. The features which make that algorithm promising are its high recognition accuracy and low computational cost, in addition to its adaptability, flexibility and applicability. Also, in a parallel line the effect of saliency as a means to generate global features is examined. In this paper, the inertial measures (instead of visual data) are utilized for person authentication. Thus the smartphone inertial sensors are used to continuously assess whether the mobile is actually in the hands of the right owner or not. Three different approaches (saliency detection, offline, and online learning) have been proposed, examined, and implemented. The last two of these approaches are based on the use of convolutional neural networks to shift the measured values of the sensors into a new vector that can be classified more reliably, while the first approach is based on the use of saliency detection algorithms to get the most salient regions of the gait. The models of these approaches are carried out and various experimentson such models are conducted. The results of these experiments were promising and showed the applicability of gait recognition to provide implicit continuous authentication, specially, when online learning is relied upon, since the identification accuracy reaches 98.7%.

GPS-assisted Indoor Pedestrian Dead Reckoning

Indoor pedestrian dead reckoning (PDR) using embedded inertial sensors in smartphones has been actively studied in the ubicomp community. However, PDR relying only on inertial sensors suffers from the accumulation of errors from the sensors. Researchers have employed various indoor landmarks detectable by smartphone sensors such as magnetic fingerprints caused by elevators and Bluetooth signals from beacons with known coordinates to compensate for the errors. This study proposes a new type of indoor landmark that does not require additional device installation, e.g., beacons, and training data collection in a target environment, e.g., magnetic fingerprints, unlike existing landmarks. This study proposes the use of GPS signals received by a smartphone to correct the accumulated errors of the PDR. While it is impossible to locate the smartphone indoors using GPS satellites, the smartphone can receive signals at a window-side area through windows from satellites aligned with the orientation of the window normal. Based on this idea, we design a machine-learning-based module for detecting the proximity of a user to a window and the orientation of the window, which enables us to roughly determine the absolute coordinates of the smartphone and to correct the accumulated errors by referring to positions of window-side areas found in the floor plan of the environment. A key technical contribution of this study is designing the module, such that it can be trained based on data from environments other than the target environment yet work in any environment by extracting GPS-related information independent of wall orientation. We evaluated the effectiveness of the proposed method using sensor data collected in real environments.

A new approach for varied speed weigh-in-motion vehicle based on smartphone inertial sensors

&lt;p&gt;Dynamic vehicle weight measuring, weigh-in-motion (WIM), is an important metric that can reflect significantly vehicle driving behaviour and in turn, it will affect both safety and traffic status. Several accurate WIM systems are developed and implemented successfully. These systems are using under road weighing sensor which are costly to implement. Moreover, it is costly and not very practical to embed a continuous weighing system in used cars. In this work, a low-cost varied-speed weigh-in-motion approach was suggested to continuously measuring vehicle load based on the response of smartphone sensors which is a reflection of vehicle dynamics. This approach can apply to any moving vehicle at any driving speed without the need for extra added hardware which makes it very applicable because smartphone is widely used device. The approach was tested through a six-trips experiment. Three capacities of load had been designed in this approach to be classified using a neural network classifier. The classification performance metrics are calculated and show an accuracy of 91.2%. This accuracy level is within error limits of existing WIM systems especially for high speed and proved the success of the suggested approach.&lt;/p&gt;

Native Smartphone Single- and Dual-Frequency GNSS-PPP/IMU Solution in Real-World Driving Scenarios

The Global Navigation Satellite System (GNSS) capability in smartphones has seen significant upgrades over the years. The latest ultra-low-cost GNSS receivers are capable of carrier-phase tracking and multi-constellation, dual-frequency signal reception. However, due to the limitations of these ultra-low-cost receivers and antennas, smartphone GNSS position solutions suffer significantly from urban multipath, poor signal reception, and signal blockage. This paper presents a novel sensor fusion technique using Precise Point Positioning (PPP) and the inertial sensors in smartphones, combined with a single- and dual-frequency (SFDF) optimisation scheme for smartphones. The smartphone is field-tested while attached to a vehicle’s dashboard and is driven in multiple real-world situations. A total of five vehicle experiments were conducted and the solutions show that SFDF-PPP outperforms single-frequency PPP (SF-PPP) and dual-frequency PPP (DF-PPP). Solutions can be further improved by integrating with native smartphone IMU measurements and provide consistent horizontal positioning accuracy of &lt;2 m rms through a variety obstructions. These results show a significant improvement from the existing literature using similar hardware in challenging environments. Future work will improve optimising inertial sensor calibration and integrate additional sensors.

Wearable Devices and Smartphone Inertial Sensors for Static Balance Assessment: A Concurrent Validity Study in Young Adult Population.

Falls represent a public health issue around the world and prevention is an important part of the politics of many countries. The standard method of evaluating balance is posturography using a force platform, which has high financial costs. Other instruments, such as portable devices and smartphones, have been evaluated as low-cost alternatives to the screening of balance control. Although smartphones and wearables have different sizes, shapes, and weights, they have been systematically validated for static balance control tasks. Different studies have applied different experimental configurations to validate the inertial measurements obtained by these devices. We aim to evaluate the concurrent validity of a smartphone and a portable device for the evaluation of static balance control in the same group of participants. Twenty-six healthy and young subjects comprised the sample. The validity for static balance control evaluation of built-in accelerometers inside portable smartphone and wearable devices was tested considering force platform recordings as a gold standard for comparisons. A linear correlation (r) between the quantitative variables obtained from the inertial sensors and the force platform was used as an indicator of the concurrent validity. Reliability of the measures was calculated using Intraclass correlation in a subsample (n = 14). Smartphones had 11 out of 12 variables with significant moderate to very high correlation (r > 0.5, p < 0.05) with force platform variables in open eyes, closed eyes, and unipedal conditions, while wearable devices had 8 out of 12 variables with moderate to very high correlation (r > 0.5, p < 0.05) with force platform variables under the same task conditions. Significant reliabilities were found in closed eye conditions for smartphones and wearables. The smartphone and wearable devices had concurrent validity for the static balance evaluation and the smartphone had better validity results than the wearables for the static balance evaluation.

Activity recognition on smartphones using an AKNN based support vectors

PurposeMobile phone-based human activity recognition (HAR) consists of inferring user’s activity type from the analysis of the inertial mobile sensor data. This paper aims to mainly introduce a new classification approach called adaptive k-nearest neighbors (AKNN) for intelligent HAR using smartphone inertial sensors with a potential real-time implementation on smartphone platform.Design/methodology/approachThe proposed method puts forward several modification on AKNN baseline by using kernel discriminant analysis for feature reduction and hybridizing weighted support vector machines and KNN to tackle imbalanced class data set.FindingsExtensive experiments on a five large scale daily activity recognition data set have been performed to demonstrate the effectiveness of the method in terms of error rate, recall, precision, F1-score and computational/memory resources, with several comparison with state-of-the art methods and other hybridization modes. The results showed that the proposed method can achieve more than 50% improvement in error rate metric and up to 5.6% in F1-score. The training phase is also shown to be reduced by a factor of six compared to baseline, which provides solid assets for smartphone implementation.Practical implicationsThis work builds a bridge to already growing work in machine learning related to learning with small data set. Besides, the availability of systems that are able to perform on flight activity recognition on smartphone will have a significant impact in the field of pervasive health care, supporting a variety of practical applications such as elderly care, ambient assisted living and remote monitoring.Originality/valueThe purpose of this study is to build and test an accurate offline model by using only a compact training data that can reduce the computational and memory complexity of the system. This provides grounds for developing new innovative hybridization modes in the context of daily activity recognition and smartphone-based implementation. This study demonstrates that the new AKNN is able to classify the data without any training step because it does not use any model for fitting and only uses memory resources to store the corresponding support vectors.