UCI Human Activity Recognition Research Articles

Feature-Level Fusion Multi-Sensor Aggregation Temporal Network for Smartphone-Based Human Activity Recognition

Smartphone-based Human Activity Recognition (HAR) identifies human movements using inertial signals gathered from multiple smartphone sensors. Generally, these signals are stacked as one (data-level fusion) and fed into deep learning algorithms for feature extractions. This research studies feature-level fusion, individually processing inertial signals from each sensor, and proposes a lightweight deep temporal learning model, Feature-Level Fusion Multi-Sensor Aggregation Temporal Network (FLF-MSATN), that performs feature extraction on inertial signals from each sensor separately. The raw signals, segmented into equally sized time windows, are passed into individual Dilated-Pooled Convolutional Heads (DPC Heads) for temporal feature analysis. Each DPC Head has a spatiotemporal block containing dilated causal convolutions and average pooling, to extract underlying patterns. The DPC Heads’ outputs are concatenated and passed into a Global Average Pooling layer to generate a condensed confidence map before activity classification. FLF-MSATN is assessed using a subject-independent protocol on a publicly available HAR dataset, UCI HAR, and a self-collected HAR dataset, achieving 96.67% and 82.70% accuracies, respectively. A Data-Level Fusion MSATN is built to compare and verify the model performance attained by the proposed FLF-MSATN. The empirical results show that implementing FLF-MSATN enhances the accuracy by ~3.4% for UCI HAR and ~9.68% for self-collected datasets.

Video Surveillance System-Based Human Activity Recognition Using Hierarchical Auto-Associative Polynomial Convolutional Neural Network with Garra Rufa Fish Optimization

Nowadays, video surveillance systems are gaining increasing attention in the computer vision fields due to user demands for security purposes. Observing human movement and predicting such movement perception are promising due to video surveillance systems. In this work, Hierarchical Auto-Associative Polynomial Convolutional Neural Networks (HA-PCNN) with Garran Rufa Fish Optimization (GRFO) is proposed for Human Activity Recognition (HAR) using a video surveillance system. Initially, video surveillance systems-based human activity images are obtained, which are collected by video camera by recording daily human activities. After that, the input images are fed to a pre-processing approach named as Switched Mode Fuzzy Median Filter (SMFM) method. In this, the noise presented in the input images is reduced by applying the SMFM model to normalize the dataset and improve the image quality. After that, the pre-processed images are given to the developed Fast Discrete Curvelet Transform with Wrapping (FDCT-WRP)-based feature extraction method to extract the relevant features. Moreover, the extracted features are given to the HA-PCNN model for HA classification. Generally, the HA-PCNN method does not express any adoption of optimization methods for scaling the ideal parameters and classification. Here, Garra Rufa Fish Optimization (GRFO) is proposed to enhance the parameters of HA-PCNN. Thus, the proposed HA-PCNN-GRFO methodology can classify human activities like standing, sitting, running, walking and sleeping. By then the performance of the proposed HA-PCNN-GRFO methodology is evaluated using Python Platform, and metrics like recall, accuracy, F-measure, precision, and specificity are analyzed. Thus, the proposed HA-PCNN-GRFO approach achieved 97.3% average accuracy rate, 96.5% average recall, 97.2% average precision, 95.9% average specificity, and 96.8% average F1-score for classifying human activities using UCI HAR dataset. Also, the proposed HA-PCNN-GRFO approach is outperformed in HAR than the conventional approaches.

Hardware-algorithm co-design of Energy Efficient Federated Learning in Quantized Neural Network

The real-world implementation of federated learning (FL) for smart device-based applications necessitates energy-efficient convergent models due to resource-constrained devices. In the literature, deep learning (DL)–based models are frequently trained on FL to obtain high accuracy and quick convergence employing a 32-bit floating point precision level. However, such circumstances are not feasible for devices with limited resources. DL models demand a significant amount of processing power and energy, which has a noticeable negative impact on the environment. Hence, it is necessary to reduce the level of precision in DL models to enhance energy efficiency. In this paper, we propose a low-precision level green federated learning (GFL) model to maintain the balance of communication cycles, energy efficiency, and accuracy with an admissible convergence rate of the deep learning algorithms. Our proposed GFL framework achieves 98.04% server accuracy and 97.69% federated accuracy with a faster convergence rate and fewer communication rounds than state-of-the-art methods on the UCI human activity recognition dataset.

A hybrid TCN-GRU model for classifying human activities using smartphone inertial signals.

Recognising human activities using smart devices has led to countless inventions in various domains like healthcare, security, sports, etc. Sensor-based human activity recognition (HAR), especially smartphone-based HAR, has become popular among the research community due to lightweight computation and user privacy protection. Deep learning models are the most preferred solutions in developing smartphone-based HAR as they can automatically capture salient and distinctive features from input signals and classify them into respective activity classes. However, in most cases, the architecture of these models needs to be deep and complex for better classification performance. Furthermore, training these models requires extensive computational resources. Hence, this research proposes a hybrid lightweight model that integrates an enhanced Temporal Convolutional Network (TCN) with Gated Recurrent Unit (GRU) layers for salient spatiotemporal feature extraction without tedious manual feature extraction. Essentially, dilations are incorporated into each convolutional kernel in the TCN-GRU model to extend the kernel's field of view without imposing additional model parameters. Moreover, fewer short filters are applied for each convolutional layer to alleviate excess parameters. Despite reducing computational cost, the proposed model utilises dilations, residual connections, and GRU layers for longer-term time dependency modelling by retaining longer implicit features of the input inertial sequences throughout training to provide sufficient information for future prediction. The performance of the TCN-GRU model is verified on two benchmark smartphone-based HAR databases, i.e., UCI HAR and UniMiB SHAR. The model attains promising accuracy in recognising human activities with 97.25% on UCI HAR and 93.51% on UniMiB SHAR. Since the current study exclusively works on the inertial signals captured by smartphones, future studies will explore the generalisation of the proposed TCN-GRU across diverse datasets, including various sensor types, to ensure its adaptability across different applications.

Research on TCN Model Based on SSARF Feature Selection in the Field of Human Behavior Recognition

Human behavior recognition is the process of automatically identifying and analyzing multiple human behaviors using modern technology. From previous studies, we find that redundant features not only slow down the model training process and increase the structural complexity but also degrade the overall performance of the model. To overcome this problem, this paper investigates a temporal convolutional neural network (TCN) model based on improved sparrow search algorithm random forest (SSARF) feature selection to accurately identify human behavioral traits based on wearable devices. The model is based on the TCN classification model and incorporates a random forest with the sparrow optimization algorithm to perform dimensionality reduction on the original features, which is used to remove poorly correlated and unimportant features and retain effective features with a certain contribution rate to generate the optimal feature subset. In order to verify the reliability of the method, the performance of the model was evaluated on two public datasets, UCI Human Activity Recognition and WISDM, respectively, and 98.54% and 97.83% recognition accuracies were obtained, which were improved by 0.47% and 1.04% compared to the prefeature selection, but the number of features was reduced by 84.31% and 32.50% compared to the original feature set. In addition, we compared the TCN classification model with other deep learning models in terms of evaluation metrics such as F1 score, recall, precision, and accuracy, and the results showed that the TCN model outperformed the other control models in all four metrics. Meanwhile, it also outperforms the existing recognition methods in terms of accuracy and other aspects, which have some practical application value.

Human activity recognition through deep learning: Leveraging unique and common feature fusion in wearable multi-sensor systems

With the progress in IoT and AI technologies, multi-sensor fusion for human activity recognition (HAR) has garnered considerable attention. As a result of integrating diverse information from different sensors, individuals employ sensors to monitor their daily activities. However, identifying crucial features for classification and assigning suitable weights to sensors is a complex task. On the other side, varied data structures pose challenges in establishing a unified format for the fusion of diverse data. To address these challenges, this paper presents UC Fusion, a method focusing on the fusion of unique and common features in wearable multi-sensor systems for HAR. First, UC Fusion merges the unique feature of each sensor with the common features of all sensors. Second, it tackles the challenge of handling heterogeneous data by unifying the data format through segmentation and dimensional transformation. Extensive experiments on the UCI HAR and WISDM datasets were conducted to evaluate UC Fusion’s performance. The results demonstrate that our proposed method secured an average recognition accuracy of 96.84% and 98.85%. Furthermore, ablation studies were performed on each module of UC Fusion to assess their impact on accuracy, confirming the effectiveness of the proposed method.

Are Gyroscopes an Added Value in Leave-One-Subject-Out Activity Recognition with IMUs?

Inertial sensors have played a key role in the development of Human Activity Recognition (HAR) systems. Adding gyroscopes in HAR systems leads to increased battery and processing resources. Therefore, it is important to explore their added value compared with using accelerometers only. This study evaluates the added value of gyroscopes in activity recognition. Two public available datasets recorded by accelerometers and gyroscopes were studied. These datasets focus on multiple types of activities: UCI HAR dataset includes walking, walking upstairs, walking downstairs, sitting, standing, laying and WISDM dataset includes 18 hand-oriented and non-hand-oriented activities. Several machine learning models were applied to both datasets for activity recognition. Leave-one-subject-out cross-validation (LOSO) was applied to evaluate the models, where the training set and test set were from different subjects. For UCI HAR dataset, the multilayer perceptron (MLP) model obtained the highest f1-scores. Adding a gyroscope on the waist significantly improved the f1-scores of sitting and laying (both ). For WISDM dataset, the support vector machines (SVM) model obtained the highest f1-scores. The gyroscope on the wrist improved hand-oriented activities while the gyroscope in the pockets improved non-hand-oriented activities (all . The results showed the improvement for recognition performance by adding gyroscopes. However, the improvement was dependent on the type of activity and the mounting place of the gyroscope. Clinical relevance- Gyroscopes are common sensors for activity recognition in wearable healthcare systems. This study proves the added value by adding gyroscopes on different mounting places for recognition performance.

Guided regularized random forest feature selection for smartphone based human activity recognition.

Human activity recognition (HAR) is an eminent area of research due to its extensive scope of applications in remote health monitoring, sports, smart home, and many more. Smartphone-based HAR systems use high-dimensional sensor data to infer human physical activities. Researchers continuously endeavor to select pertinent and non-redundant features without compromising the classification accuracy. In this work, our aim is to build an efficient HAR model that not only extracts the most relevant features from the 3-axial accelerometer and gyroscope signal data but also enhances the classification accuracy of the HAR system, without data loss using time-frequency domain features. We propose a feature selection method based on guided regularized random forest (GRRF) to determine the most pertinent and non-redundant features to reduce the time to recognize the human activities efficiently. After selecting the most relevant features, a support vector machine (SVM) is used to identify various human physical activities. The UCI public dataset and a self-collected dataset are used to assess the generalization capability and performance of the proposed feature selection method. Eventually, the accuracy reached 99.10% and 99.30% on the self-collected and UCI HAR datasets, respectively.

A Novel CNN-based Bi-LSTM parallel model with attention mechanism for human activity recognition with noisy data

Boosted by mobile communication technologies, Human Activity Recognition (HAR) based on smartphones has attracted more and more attentions of researchers. One of the main challenges is the classification time and accuracy in processing long-time dependent sequence samples with noisy or missed data. In this paper, a 1-D Convolution Neural Network (CNN)-based bi-directional Long Short-Term Memory (LSTM) parallel model with attention mechanism (ConvBLSTM-PMwA) is proposed. The original features of sensors are segmented into sub-segments by well-designed equal time step sliding window, and fed into 1-D CNN-based bi-directional LSTM parallel layer to accelerate feature extraction with noisy and missed data. The weights of extracted features are redistributed by attention mechanism and integrated into complete features. At last, the final classification results are obtained with the full connection layer. The performance is evaluated on public UCI and WISDM HAR datasets. The results show that the ConvBLSTM-PMwA model performs better than the existing CNN and RNN models in both classification accuracy (96.71%) and computational time complexity (1.1 times faster at least), even if facing HAR data with noise.

Tuning the Hyperparameters of the 1D CNN Model to Improve the Performance of Human Activity Recognition

The human activity recognition (HAR) field has recently become one of the trendiest research topics due to ready-made sensors such as accelerometers and gyroscopes equipped with smartphones and smartwatches as an embedded devices, decreasing the cost and power consumption. As a result, human activity is considered a time series classification problem. Now a day, deep learning approaches such as Convolutional Neural Network (CNN) have been successful when implemented with HAR to learn automatically higher-order features and, at the same time, work as a classifier. Recently, a one-dimensional Convolutional Neural Network (1D CNN) has been suggested and carried out at the best performance levels in numerous applications, such as the classification of personalized biomedical data and time series classification. This paper studies how to leverage a 1D single CNN model to produce an excellent performance on the human activity raw data. This is done by empirically tuning the values of hyperparameters, such as kernel size, filter maps, number of epochs, batch size, and promoting an advanced multi-headed 1D CNN by employing each convolutional layer with a different kernel size to gain an ensemble–like results. The selected hyper parameter's impact is evaluated on a publicly available dataset named UCI HAR collected from smartphone sensors to perform six activities. A significant determinant of better results depends on the hyperparameter that has been chosen. The results demonstrated that tuning the hyperparameter of 1D CNN increased activity recognition accuracy.

Comparative analysis of HAR datasets using classification algorithms

In the current research and development era, Human Activity Recognition (HAR) plays a vital role in analyzing the movements and activities of a human being. The main objective of HAR is to infer the current behaviour by extracting previous information. Now-a-days, the continuous improvement of living condition of human beings changes human society dramatically. To detect the activities of human beings, various devices, such as smartphones and smart watches, use different types of sensors, such as multi modal sensors, non-video based and video-based sensors, and so on. Among the entire machine learning approaches, tasks in different applications adopt extensively classification techniques, in terms of smart homes by active and assisted living, healthcare, security and surveillance, making decisions, tele-immersion, forecasting weather, official tasks, and prediction of risk analysis in society. In this paper, we perform three classification algorithms, Sequential Minimal Optimization (SMO), Random Forest (RF), and Simple Logistic (SL) with the two HAR datasets, UCI HAR and WISDM, downloaded from the UCI repository. The experiment described in this paper uses the WEKA tool to evaluate performance with the matrices, Kappa statistics, relative absolute error, mean absolute error, ROC Area, and PRC Area by 10-fold cross validation technique. We also provide a comparative analysis of the classification algorithms with the two determined datasets by calculating the accuracy with precision, recall, and F-measure metrics. In the experimental results, all the three algorithms with the UCI HAR datasets achieve nearly the same accuracy of 98%.The RF algorithm with the WISDM dataset has the accuracy of 90.69%,better than the others.

Comparing Human Activity Recognition Models Based on Complexity and Resource Usage

Human Activity Recognition (HAR) is a field with many contrasting application domains, from medical applications to ambient assisted living and sports applications. With ever-changing use cases and devices also comes a need for newer and better HAR approaches. Machine learning has long been one of the predominant techniques to recognize activities from extracted features. With the advent of deep learning techniques that push state of the art results in many different domains like natural language processing or computer vision, researchers have also started to build deep neural nets for HAR. With this increase in complexity, there also comes a necessity to compare the newer approaches to the previous state of the art algorithms. Not everything that is new is also better. Therefore, this paper aims to compare typical machine learning models like a Random Forest (RF) or a Support Vector Machine (SVM) to two commonly used deep neural net architectures, Convolutional Neural Nets (CNNs) and Recurrent Neural Nets (RNNs). Not only in regards to performance but also in regards to the complexity of the models. We measure complexity as the memory consumption, the mean prediction time and the number of trainable parameters of the models. To achieve comparable results, the models are all tested on the same publicly available dataset, the UCI HAR Smartphone dataset. With this combination of prediction performance and model complexity, we look for the models achieving the best possible performance/complexity tradeoff and therefore being the most favourable to be used in an application. According to our findings, the best model for a strictly memory limited use case is the Random Forest with an F1-Score of 88.34%, memory consumption of only 0.1 MB and mean prediction time of 0.22 ms. The overall best model in terms of complexity and performance is the SVM with a linear kernel with an F1-Score of 95.62%, memory consumption of 2 MB and a mean prediction time of 0.47 ms. The two deep neural nets are on par in terms of performance, but their increased complexity makes them less favourable to be used.

Optimization of deep neural network-based human activity recognition for a wearable device

Human activity recognition (HAR) attempts to classify performed activities from data retrieved from different sensors attached to the body. Most publications pertaining to HAR based on deep neural networks (DNNs) report the development of a suitable architecture to improve recognition accuracy by increasing the parameters of the architecture. Our work follows a different approach by attempting to optimise DNN-based HAR by reducing the dimensions of acceleration data, by finding a suitable sample size for processing by the DNN and by reducing the parameters of the proposed architecture. The experiments rely on employing two previously presented DNN-based HAR architectures as the baselines and starting points to create our candidate architectures. The variations in the dimensions of acceleration data, i.e., {xy, yz, xz, x, y, z}, and the sample size, i.e. {4, 6, 8} s duration, to these candidate architectures are experimented to produce the winner architecture which takes the shortest sample size and the minimal dimensions of acceleration data while preserving the recognition precision. The results indicate that despite the number of parameters is approximately half of the baseline architecture with two dimensions of acceleration data and shorter sample size (i.e., using a sample of 4 s duration instead of 8 s and only the xy axes of acceleration data), the resulting DNN-based HAR classifiers can produce comparable or better recognition precision than the baseline classifiers. The experimental results were obtained using three different popular datasets: the WISDM, the UCI HAR, and the Real World 2016. The proposed classifiers with optimised settings are useful as they require less processing time and reduce power consumption, both in terms of retrieving acceleration data from the sensor and the CPU processing time. Furthermore, they reduce the memory requirements for parameter storing and are suitable for incorporation in a wearable device.

Attention induced multi-head convolutional neural network for human activity recognition

Deep neural networks, including convolutional neural networks (CNNs), have been widely adopted for human activity recognition in recent years. They have attained significant performance improvement over traditional techniques due to their strong feature representation capabilities. Some of the challenges faced by the HAR community is the non-availability of a substantial amount of labeled training samples, and the higher computational cost and system resources requirements of deep learning architectures as opposed to shallow learning algorithms. To address these challenges, we propose an attention-based multi-head model for human activity recognition (HAR). This framework contains three lightweight convolutional heads, with each head designed using one-dimensional CNN to extract features from sensory data. The lightweight multi-head model is induced with attention to strengthen the representation ability of CNN, allowing for automatic selection of salient features and suppress unimportant ones. We conducted ablation studies and experiments on two publicly available benchmark datasets: WISDM and UCI HAR, to evaluate our model. The experimental outcome demonstrates the effectiveness of the proposed framework in activity recognition and achieves better accuracy while ensuring computational efficiency.

Hybrid deep learning approaches for smartphone sensor-based human activity recognition

Human Activity Recognition (HAR) has become one of the most important research fields to achieve real-time monitoring of human activities for timely decision making in various applications like fall detection, elderly care etc. Now-a-days, most people use smartphones which come with various embedded inertial sensors like accelerometer and gyroscope to monitor acceleration and angular velocity. These smartphone-based sensors have proven to be cost-effective solution in identification of activities belonging to ADL (Activities of Daily Living). Various Machine Learning, Deep learning and hybrid models have been proposed and implemented for HAR. This paper also proposes various hybrid deep learning approaches which combine Deep Neural Networks with other models like LSTM (Long Short Term Memory) Model and GRU (Gated Recurrent Unit) for effective classification of engineered features from CNN (Convolutional Neural Network) Model. A novel architecture that integrates CNN with Random Forest Classifier (DeepCNN-RF) is proposed to add randomness to the model. The proposed models have been tested on publicly available HAR Datasets like UCI HAR and WISDM Activity Recognition Datasets. Experimental results show that the hybrid models outperform the state-of-the-art data mining, machine learning techniques in UCI HAR and WISDM with an overall maximum accuracy of 97.77% and 98.2% respectively.

The Layer-Wise Training Convolutional Neural Networks Using Local Loss for Sensor-Based Human Activity Recognition

Recently, deep learning, which are able to extract automatically features from data, has achieved state-of-the-art performance across a variety of sensor based human activity recognition (HAR) tasks. However, the existing deep neural networks are usually trained with a global loss, and all hidden layer weights have to be always kept in memory before the forward and backward pass has completed. The backward locking phenomenon prevents the reuse of memory, which is a crucial limitation for wearable activity recognition. In the paper, we proposed a layer-wise convolutional neural networks (CNN) with local loss for the use of HAR task. To our knowledge, this paper is the first that uses local loss based CNN for HAR in ubiquitous and wearable computing arena. We performed experiments on five public HAR datasets including UCI HAR dataset, OPPOTUNITY dataset, UniMib-SHAR dataset, PAMAP dataset, and WISDM dataset. The results show that local loss works better than global loss for tested baseline architectures. At no extra cost, the local loss can approach the state-of-the-arts on a variety of HAR datasets, even though the number of parameters was smaller. We believe that the layer-wise CNN with local loss can be used to update the existing deep HAR methods.

Highly efficient federated learning with strong privacy preservation in cloud computing

Federated learning is a new machine learning framework that allows mutually distrusting clients to reap the benefits from the joint training model without explicitly disclosing their private datasets. However, the high communication cost between the cloud server and clients has become the main challenge due to the limited network bandwidth. Moreover, the model parameters it shares may be utilized to perform model inversion attacks. Aimed at these problems, a new scheme for highly efficient federated learning with strong privacy preservation in cloud computing is presented. We design a lightweight encryption protocol to provide provably privacy preservation while maintaining desirable model utility. Additionally, an efficient optimization strategy is employed to enhance the training efficiency. Under the defined threat model, we prove the proposed scheme is secure against the honest-but-curious server and extreme collusion. We evaluate the effectiveness of our scheme and compare it with existing related works on MNIST and UCI Human Activity Recognition Dataset. Results show that our scheme reduces the execution time by 20% and transmitted ciphertext size by 85% on average while achieving similar accuracy as the compared secure multiparty computation (SMC) based methods.

Human action recognition with deep learning and structural optimization using a hybrid heuristic algorithm

Human action recognition (HAR) is a popular subject for academic society and other stakeholders. Nowadays it has a wide-spread use for lots of practical applications such as for health, assistive living, elderly care, and so on. Both visual and sensor-based data can be used for HAR. Visual data includes video images, still images, skeleton images, etc., whilst sensor-based data is acquired as numerical data from devices such as accelerometer, gyroscope, and so on. Employed classification methods and data types are of crucial importance on HAR performance. In this paper, sensor data based activity recognition is performed using stacked autoencoders (SAE). Finding near optimal accuracy results with SAEs is a challenging process if structural optimization is left to user experience. The purpose of this study is to improve the accuracy of HAR classifying methods such as SAEs using heuristic optimization algorithms. Hence the structural parameters of SAEs have been optimized using artificial bee colony optimization algorithm (ABC), genetic algorithm, differential evolution algorithm, particle swarm optimization algorithm (PSO) and an afresh developed hybrid algorithm (hABCPSO) which includes PSO and ABC in its internal structure. Leave-one-out cross-validation (LOOCV) test method is used for validating the results. Each algorithm is performed for 30 runs and the results of these runs are analyzed by statistical methods in detail. According to experimental results, hABCPSO supported SAE gives the minimum error and is the most robustness algorithm among the others. Obtained success rates show that the proposed SAE achieved the best accuracy rate ever on UCI human activity recognition dataset and a close result to best on wireless sensor data mining dataset with regard to LOOCV test technique.

Human Activity Recognition Based on Gramian Angular Field and Deep Convolutional Neural Network

With the development of the Internet of things (IoT) and wearable devices, the sensor-based human activity recognition (HAR) has attracted more and more attentions from researchers due to its outstanding characteristics of convenience and privacy. Meanwhile, deep learning algorithms can extract high-dimensional features automatically, which makes it possible to achieve the end-to-end learning. Especially the convolutional neural network (CNN) has been widely used in the field of computer vision, while the influence of environmental background, camera shielding, and other factors are the biggest challenges to it. However, the sensor-based HAR can circumvent these problems well. Two improved HAR methods based on Gramian angular field (GAF) and deep CNN are proposed in this paper. Firstly, the GAF algorithm is used to transform the one-dimensional sensor data into the two-dimensional images. Then, through the multi-dilated kernel residual (Mdk-Res) module, a new improved deep CNN network Mdk-ResNet is proposed, which extracts the features among sampling points with different intervals. Furthermore, the Fusion-Mdk-ResNet is adopted to process and fuse data collected by different sensors automatically. The comparative experiments are conducted on three public activity datasets, which are WISDM, UCI HAR and OPPORTUNITY. The optimal results are obtained by using the indexes such as accuracy, precision, recall and F-measure, which verifies the effectiveness of the proposed methods.

A kernel semi-supervised distance metric learning with relative distance: Integration with a MOO approach

Metric learning, which aims to determine an appropriate distance function to measure the similarity and dissimilarity between data points accurately, is one of the most popular methods to enhance the performance of many machine learning methods such as K-means clustering and K nearest neighbor classifier algorithms. These algorithms may not perform well because of the use of normal Euclidean distance function that ignores any statistical regularities that might be estimated from a large training set of labeled examples. In many real-world applications, the Euclidean distance may not be fit to capture the intrinsic similarity and dissimilarity between the data points. Compared to existing metric learning algorithms, which use large amount of labeled data in the form of must-link (ML) and cannot-link constraints as side information where the granularity of true clustering is unknown, our proposed approach uses few labeled data in the form of relative-distance constraints such as equality constraints, Ceq, and inequality constraints, Cneq. For satisfying such constraints, we need to project the initial Euclidean distance matrix by using Bregman projection on the convex subset of constraints in such a way that all the constraints are satisfied. Since Bregman projection is not orthogonal, means while satisfying the current constraint previously satisfied constraints may get unsatisfied, we need to select a proper subset of constraints for learning better distance function. The multi-objective framework is utilized for selecting a good subset of constraints which can help in getting the proper labeling of the data set. The selected subset of constraints is used for adjusting the initial kernel-matrix. K-means clustering technique is applied to the adjusted kernel matrix to label the data set. In order to evaluate the quality of obtained labeling, different external and internal cluster validity indices are deployed. The values of these indices are simultaneously optimized using the search capability of MOO with the aim of selecting the appropriate subset of constraints. The proposed approach is evaluated on UCI Human Activity Recognition using Smartphone Dataset v1.0 along with nine other popular data sets. Results show that our approach outperforms the state of the art semi-supervised metric learning algorithms with respect to different internal and external cluster validity indices.