Human Activity Recognition Using WISDM: Exploring Class Balancing and ML Techniques

EasyChair Preprint no. 13545

Abstract

Wearable sensor for human activity recognition (HAR) is vital in activity sensing research. We addressed dataset imbalance in WISDM using three class balancing techniques: SMOTE, LoRAS, and ProWRA, applied to five machine learning models. LoRAS consistently achieved the highest accuracy, recall, precision, and F1-scores, outperforming SMOTE and ProWRA. Our results demonstrate LoRAS as the most effective technique for enhancing model performance in human activity recognition.

Keyphrases: LoRAS, ProWRA, SMOTE

@Booklet{EasyChair:13545,
  author = {Maria Hanif and Rizwan Ahmad and Shams Qazi and Waqas Ahmed and Muhammad Mahtab Alam},
  title = {Human Activity Recognition Using WISDM: Exploring Class Balancing and ML Techniques},
  howpublished = {EasyChair Preprint no. 13545},

  year = {EasyChair, 2024}}