The ubiquity of smartphones and the growth of computing resources, such as connectivity, processing, portability, and power of sensing, have greatly changed people’s lives. Today, many smartphones contain a variety of powerful sensors, including motion, location, network, and direction sensors. Motion or inertial sensors (e.g., accelerometer), specifically, have been widely used to recognize users’ physical activities. This has opened doors for many different and interesting applications in several areas, such as health and transportation. In this perspective, this work provides a comprehensive, state of the art review of the current situation of human activity recognition (HAR) solutions in the context of inertial sensors in smartphones. This article begins by discussing the concepts of human activities along with the complete historical events, focused on smartphones, which shows the evolution of the area in the last two decades. Next, we present a detailed description of the HAR methodology, focusing on the presentation of the steps of HAR solutions in the context of inertial sensors. For each step, we cite the main references that use the best implementation practices suggested by the scientific community. Finally, we present the main results about HAR solutions from the perspective of the inertial sensors embedded in smartphones.

In this article, we introduce explainable methods to understand how Human Activity Recognition (HAR) mobile systems perform based on the chosen validation strategies. Our results introduce a new way to discover potential bias problems that overestimate the prediction accuracy of an algorithm because of the inappropriate choice of validation methodology. We show how the SHAP (Shapley additive explanations) framework, used in literature to explain the predictions of any machine learning model, presents itself as a tool that can provide graphical insights into how human activity recognition models achieve their results. Now it is possible to analyze which features are important to a HAR system in each validation methodology in a simplified way. We not only demonstrate that the validation procedure k-folds cross-validation (k-CV), used in most works to evaluate the expected error in a HAR system, can overestimate by about 13% the prediction accuracy in three public datasets but also choose a different feature set when compared with the universal model. Combining explainable methods with machine learning algorithms has the potential to help new researchers look inside the decisions of the machine learning algorithms, avoiding most times the overestimation of prediction accuracy, understanding relations between features, and finding bias before deploying the system in real-world scenarios.

This work proposes a new indoor positioning system, named KLIP, that uses the ${K}$ -means clustering algorithm to split the environment into different sets of log-distance propagation models in order to better characterize the indoor environment and further improve the position estimation using Bayesian inference. Our proposed method is validated in a large-scale, real-world scenario composed of Bluetooth Low Energy (BLE)-based devices. We demonstrate, throughout the paper, that the addition of location information of training points to the received signal strength indicator (RSSI) as an attribute for the clustering step improves the positioning accuracy. Moreover, the obtained results show that our solution outperforms the naive Bayesian estimation up to 12% – regarding the positioning accuracy – and the broadly deployed ${k}$ NN for reduced training dataset size – regarding both accuracy and online processing time. In this sense, KLIP proves to be an efficient and scalable alternative when both site-survey effort and energy consumption constraints must be taken into account.

Indoor positioning systems (IPSs) allow the location and tracking of mobile devices in indoor environments where the global positioning system (GPS) does not provide satisfactory results. In model-based IPSs, it is common to use signal propagation models to estimate distances between anchor nodes and mobile devices using the received signal strength indicator (RSSI). However, using fixed parameters in the path loss model to characterize the signal in large-scale scenarios results in the degradation of the positioning accuracy. In this article, we propose the adaptive model (ADAM) positioning system, a model-based IPS that chooses the best anchor nodes to benefit the positioning computation and uses different parameters for the log-distance model to represent the signal in different regions and conditions of the scenario. Then, we estimate a single, more precise position using a data fusion technique. Our proposal does not require training nor prior knowledge of the best parameters for each region. We evaluated the performance of our proposed system in a real-world, large-scale environment using Bluetooth-based mobile devices. Our results clearly show that ADAM can locate mobile devices with an average error of 2.93 m in relation to the real position, which is 23% better than literature-based models using fixed parameters for the entire environment.

The use of machine learning (ML) techniques in affective computing applications focuses on improving the user experience in emotion recognition. The collection of input data (e.g., physiological signals), together with expert annotations are part of the established standard supervised learning methodology used to train human emotion recognition models. However, these models generally require large amounts of labeled data, which is expensive and impractical in the healthcare context, in which data annotation requires even more expert knowledge. To address this problem, this paper explores the use of the self-supervised learning (SSL) paradigm in the development of emotion recognition methods. This approach makes it possible to learn representations directly from unlabeled signals and subsequently use them to classify affective states. This paper presents the key concepts of emotions and how SSL methods can be applied to recognize affective states. We experimentally analyze and compare self-supervised and fully supervised training of a convolutional neural network designed to recognize emotions. The experimental results using three emotion datasets demonstrate that self-supervised representations can learn widely useful features that improve data efficiency, are widely transferable, are competitive when compared to their fully supervised counterparts, and do not require the data to be labeled for learning.

Indoor Positioning Systems (IPSs) have emerged as a research topic in mobile computing, enabling the tracking and location of mobile devices within indoor environments. In model-based IPSs, the Received Signal Strength Indicator (RSSI) is used to estimate the distance between wireless signal receivers and transmitters using signal propagation models. However, the indoor environment presents challenges that make distance estimation using RSSI difficult. In this paper, we propose a new indoor positioning system that combines Particle Swarm Optimization (PSO) with signal propagation models to improve the accuracy of mobile device positioning. The PSO algorithm is used to optimize the position estimation process by generating different particles in the map, while the signal propagation model is used to model the attenuation and reflection of wireless signals in each particle. Our MIPS-PSO system does not require any prior training nor any knowledge of the best parameters of the signal propagation model. We evaluated the performance of our system using data collected in a real indoor environment with Bluetooth-Low-Energy (BLE) devices. Our results show that the MIPS-PSO achieves an average error of 2.57 m, an improvement of 40% when compared to a traditional trilateration, model-based IPS.