Abstract:Human Activity Recognition (HAR), based on machine and deep learning algorithms is considered one of the most promising technologies to monitor professional and daily life activities for different categories of people (e.g., athletes, elderly, kids, employers) in order to provide a variety of services related, for example to well-being, empowering of technical performances, prevention of risky situation, and educational purposes. However, the analysis of the effectiveness and the efficiency of HAR methodologies suffers from the lack of a standard workflow, which might represent the baseline for the estimation of the quality of the developed pattern recognition models. This makes the comparison among different approaches a challenging task. In addition, researchers can make mistakes that, when not detected, definitely affect the achieved results. To mitigate such issues, this paper proposes an open-source automatic and highly configurable framework, named B-HAR, for the definition, standardization, and development of a baseline framework in order to evaluate and compare HAR methodologies. It implements the most popular data processing methods for data preparation and the most commonly used machine and deep learning pattern recognition models.

What problem does this paper attempt to address?

The paper primarily addresses the issues present in the field of Human Activity Recognition (HAR) research and proposes solutions. Specifically, the paper addresses the following problems: 1. **Lack of standardized workflow**: Currently, in the HAR field, there is a lack of a standardized workflow to serve as a baseline for evaluating the effectiveness and efficiency of different HAR methods, making it difficult to compare different methods. 2. **Common errors in research**: Researchers may make some common errors when implementing HAR workflows, which can affect the accuracy of the results. For example: - Incorrect handling of missing values, infinite values, or corrupted values in the data. - Incorrect application of the normalization process. - Errors in the training and testing process. - Using accuracy alone to evaluate performance in multi-class problems without considering sensitivity, specificity, precision, etc. 3. **Non-standardized steps in data preprocessing and model selection stages**: These steps are crucial for the quality of HAR methods, but there are currently no clear rules to guide how these steps should be performed. To address the above issues, the paper proposes an open-source framework named B-HAR (Baseline-HAR). The main features of this framework include: - **Unified workflow**: Integrates the typical HAR workflow shown in Figure 1 into a single framework, making various preprocessing steps transparent to the user. - **Defined sequence of data preprocessing steps**: Defines the execution order of data preprocessing steps in a way that minimizes errors. - **Provision of benchmark information**: Provides users with information about the most well-known HAR datasets, allowing new HAR methods to be compared in detail with existing results. - **Automatic application of grid search technique**: Automatically applies the grid search technique to commonly used pattern recognition models for newly introduced HAR datasets, providing baseline results to users. - **Support for user-defined models**: Allows users to implement their own HAR models (based on machine learning or deep learning) and integrate them into B-HAR while keeping the data preprocessing steps unchanged. In summary, the paper aims to simplify the setup and configuration of HAR workflows and support researchers in comparing existing and new recognition models without making errors by providing a unified and customizable framework—B-HAR.

B-HAR: an open-source baseline framework for in depth study of human activity recognition datasets and workflows

Continual learning in sensor-based human activity recognition: An empirical benchmark analysis

HARTH: A Human Activity Recognition Dataset for Machine Learning

HARE: Unifying the Human Activity Recognition Engineering Workflow

ActivBench: Leveraging Human Activity Inference from Smartphone Sensors for Human Computer Interactions

Attention-Based Deep Learning Framework for Human Activity Recognition with User Adaptation

Human Activity Recognition using Inertial, Physiological and Environmental Sensors: a Comprehensive Survey

Adaptive Feature Processing for Robust Human Activity Recognition on a Novel Multi-Modal Dataset

Bimodal HAR-An Efficient Approach to Human Activity Analysis and Recognition Using Bimodal Hybrid Classifiers.

A Comprehensive Methodological Survey of Human Activity Recognition Across Divers Data Modalities

Online Continual Learning for Human Activity Recognition

Overview of Human Activity Recognition Using Sensor Data

Comprehensive machine and deep learning analysis of sensor-based human activity recognition

Human activity recognition with smartphone-integrated sensors: A survey

A Public Domain Dataset for Human Activity Recognition in Free-Living Conditions

HARMamba: Efficient and Lightweight Wearable Sensor Human Activity Recognition Based on Bidirectional Mamba

RecurrentHAR: A Novel Transfer Learning-Based Deep Learning Model for Sequential, Complex, Concurrent, Interleaved, and Heterogeneous Type Human Activity Recognition

Human Activity Recognition (HAR) in Smart Homes

Maintenance Required: Updating and Extending Bootstrapped Human Activity Recognition Systems for Smart Homes

A Novel Framework Based on Deep Learning Architecture for Continuous Human Activity Recognition with Inertial Sensors

HADE: Exploiting Human Action Recognition Through Fine-Tuned Deep Learning Methods