Baselines for Human Activity Recognition

This repo will implement a series of baselines for Human Activity Recognition, including two main technique route:

• Manual Feature extraction + Classifier
• Edge2Edge Deeplearning + Classifier(normally mlp with sofrmax)

Here are the datasets we use:

• UCI HAR, offical web, paper.
• mHealth, offical web, paper.
• PAMAP2, offical web, paper.
• OPPORTUNITY, offical web, paper.
• WISDM, offical web, paper.
• USC HAD, offical web, paper.
• skoda, offical web, paper.

# Route 1

• Classifiers
  • Decision Tree
  • Random Forest
  • MLP
  • Adaboost with DT
  • XGBoost
  • LightGBM
  • SVC

# Route 2

• Classic Models
  • MLP
  • CNN 1d
  • LSTM
  • GRU
  • BiLSTM
  • BiGRU
• Sota Models
  • DeepConvLSTM
  • Res BiLSTM
  • TCN
  • Dilated TCN
  • CNN BiLSTM
  • CNN+LSTM+Self Attension

Reference