Table 1 Existing studies on HAR using DL, sensor-based iot, and the Edge–Cloud infrastructure to enhance accuracy, real-time processing, and system scalability.

Wazwaz et al.¹¹

To develop a dynamic and distributed HAR system using smart IoT devices, edge computing, and cloud resources for accurate and efficient real-time activity detection

Wearable Sensors and Smartphones, Accelerometer and Gyroscope Features, LightGBM

UCI HAR, WISDM, ActivityNet, and other Datasets

Accuracy, Prediction Time

Arokiaraj and Viswanathan¹²

To develop an effective HAR system for improved classification accuracy with efficient computation

CN, MGRU, ELM

Real-Time IoT Data, WISDM Dataset

Accuracy, Precision, AUC, Recall, Specificity and F1-Score

Mohsin et al.¹³

To develop an automatic HAR using multi-sensor smartphone data and DBNs for accurate behavior analysis in real-time

Multi-Sensor Data Collection Via Smartphones, Statistical, DBN

HAR Dataset

Accuracy, Real-Time Monitoring

Hafeez et al.¹⁴

To develop a hybrid feature-based HAR integrating inertial and visual data for accurate detection of human actions and falls

Background Removal and Silhouette Extraction, Human Skeleton Keypoint Extraction, Time-Frequency and Geometric Extraction, Feature Fusion and Zero-Order Optimization, LR

UP Fall, University of Rzeszow Fall, SisFall

Accuracy, Precision, Sensitivity, F1-Score

Bouazizi et al.¹⁵

To develop an accurate, non-intrusive HAR model for elderly monitoring using multiple 2D Lidar sensors and DL

Deploy Multiple 2D Lidar Sensors, Data Concatenation and Image Transformation, LSTM Neural Network

Indoor Lidar Sensor Data

Accuracy, Fall Detection

Alonazi et al.¹⁶

To develop a DL model for accurate HAR to assess cognitive health

SOSDCNN-HAR, WF, RetinaNet, GRU

Penn Action, NW-UCLA

Precision, Accuracy

Djenouri et al.¹⁷

To develop a hybrid CNN-Transformer model for robust home-based monitoring

STVL-HM, CNN

Kinetics-400

Accuracy, Robustness

Yadav et al. ¹⁸

To enhance accuracy and efficiency in HAR

VGG16, VGG19, MobileNetV2, SVM, RF, MLP, Fuzzy Classifier

KTH, YouTube11, and Peliculas

Accuracy, Precision, Recall, F1-Score

Kumar and Sumathi¹⁹

To enhance HAR accuracy using IMU spectrogram analysis with a lightweight DL and optimization approach

IMU Data to Spectrogram Via STFT, GNet, FHO, Time-Frequency and Spatial Feature Analysis

WISDM (Smartwatch, Smartphone), MOTION SENSE, UCI-HAR

Accuracy, Precision, Recall, F1-Score

Thukral et al. ²⁰

To improve sensor-based HAR in low-label scenarios using TL

TL, Teacher-Student Self-Training Framework, Cross-Domain Adaptation (Sensor Type, Location, Activity)

Seven Benchmark HAR Datasets

Accuracy, Precision, Recall, F1-Score

Sharen et al.²¹

To improve multi-class HAR accuracy using a custom 1D-CNN model with advanced feature extraction blocks

WISNet (custom 1D-CNN), CNPM, IDBN, and CASb Block

HAR, UCI-HAR, KU-HAR, Sleep Detection, Fall Detection, ECG Heartbeat

Accuracy, F1-Score

Sassi Hidri et al.²²

To improve HAR accuracy by optimizing DL models for smartphone accelerometer data

CNN-based Autoencoders, LSTM-RNN

WISDM

Accuracy, Classification Performance

Mohsen²³

To accurately classify human activities using a DL method based on GRU

GRU-based DL, TensorFlow Framework

WISDM

Accuracy, Precision, F1-Score

Mazhar et al.²⁴

To enhance IoT device security by using ML and DL for detecting cyberattack patterns in unstructured data.

AI-driven IoT Security, ML, DL

Not Specifically Mentioned (General IoT Data)

Detection Rate, Accuracy, Security Effectiveness

Rajeswari et al.²⁵

To develop a real-time, accurate, and privacy-aware health monitoring system using IoT, CNN models, and edge computing

Wearable Sensor Data Collection, CNN, Edge Computing

Real Patient Dataset

Accuracy, Real-Time Detection

Ma et al. ²⁶

To enhance badminton action recognition accuracy and efficiency using optimized DL methods for real-time training and match analysis

ST-GCN, VATRM, MM-Net

VideoBadminton Dataset

Accuracy, F1-Score, Inference Time

Ghadi et al.²⁷

To explore the integration of FL with IoT for decentralized, privacy-preserving AI in real-world applications

FL, Decentralized AI Training

Not Specified (Various IoT Data)

Privacy, Security, Model Accuracy