Skip to main content

Table 3 Non-exhaustive List of Algorithms used for multi-sensor data modelling

From: Artificial Intelligence Based Body Sensor Network Framework—Narrative Review: Proposing an End-to-End Framework using Wearable Sensors, Real-Time Location Systems and Artificial Intelligence/Machine Learning Algorithms for Data Collection, Data Mining and Knowledge Discovery in Sports and Healthcare

Algorithm

Task

Performance

Category

References

Kernel Ensemble Random Forest classifier with 40 estimators, 8 features at depth of 15

Heart disease prediction using daily activity data from multiple sensors

98% accuracy on testing data

Medical Data

[18]

Convolutional Neural Networks

Fault diagnosis in a planetary gearbox from multi-sensor data

93% to 99% accuracy on testing data

Machine Design

[58]

Long Short-Term Memory Artificial Neural Network

Real-time identification of foot contact and foot off by analyzing gait pattern in children

 ~ 95% with maximum delay of 3 s in real time

Human Motion Analysis

[59]

TimeNet Pre-trained Deep Recurrent Neural Network

Generalized time series classification across multiple datasets

The average accuracy observed was 83% on various datasets

Generalized solution for series analysis across various domains

[60]

Choquet Integral + Hidden Markov Chain Models

Multivariate Time Series Anomaly Detection across various data sets

Between 90 to 99% depending on the chosen dataset

Anomaly detection

[61]

Convolutional Neural Networks

Real-Time Skeletal Posture estimation using mm-wave radar

Localization error of 3.2 cm for X, 2.7 for y and 7.5 for z

Human Motion Capture

[39]

Principle Component Analysis + Toeplitz Inverse-Covariance Clustering

Multivariate Time series analysis for identification of recurring events in smart manufacturing

Performs best across multiple performance matrices (F1, Precision, Rand Index, etc.)

Automatic Event Detection

[62]

K-nearest neighbors

Method for Recognition of the Physical Activity of Human Being Using a Wearable Accelerometer

78.9% accuracy

Activity Recognition

[63]

Support Vector Machines

Fall detection on mobile phones using features from a five-phase mode

Recall 90% and precision 95.7%

Activity Recognition /Fall detection

[64]

Artificial neural networks

An alternative to traditional fall detection methods

Sensitivity 0.984 Specificity 0.986

Activity Recognition /Fall detection

[65]

Bayesian sequential analysis and Multilayer Perceptron

Contamination Event Detection with Multivariate Time-Series Data in Agricultural Water Monitoring

Average detection rate > 80%

Water Contamination Event Detection

[66]

Fisher's Linear Discriminant

Detecting Stress During Real-World Driving Tasks Using Physiological Sensors

Accuracy of 97.4%

Stress Level Detection

[67]

Correlation-based feature selection with random forest classifier with random forest classifier

Automated epileptic seizure detection

Average accuracy of 98.45%

Medical Diagnosis

[68]