Human Activity Recognition Data Science Project

Overview

This repository is dedicated to the Human Activity Recognition (HAR) project, which aims to classify different human activities based on the data from wearable sensors. The project focuses on developing models that can accurately recognize and classify activities like walking, reading book, using phone, writing etc.

Files

CNN

• CNN.py - CNN feature extactor model
• cnn_utils.py - util functions for CNN feature extractor

LSTM

• lstm_autoencoder.py - LSTM autoencoder model
• lstm_autoencoders_utils.py - util functionss for LSTM autoencoder

main_models - contains different experiments conducted

• cnn_to_rf.ipynb - 3D CNN feature extractor to Random Forest model
• cnn_to_xgb.ipynb - 3D CNN feature extractor to XGBoost model
• embedding_nn.ipynb - LSTM autoencoder to Neural Network model
• embedding_rf.ipynb - LSTM autoencoder to Random Forest model
• lstm+cnn_rf.ipynb - LSTM+CNN feature extractor to Random Forest model
• lstm_secret_data.ipynb - 3D CNN feature extractor on extended train data with filled gaps into LSTM model
• only_1Dcnn.ipynb - only 1D CNN feature extractor model
• only_cnn.ipynb - only 3D CNN feature extractor model
• only_rf.ipynb - only Random Forest model
• only_xgboost.ipynb - only XGBoost model
• simple_prob.ipynb - probability according to classes distribution in train

main_utils - containts utils for main

• fill_ranges_script.ipynb - fills ranges in train_data.csv to extedn train data
• generate_graphs.ipynb - generate graphs from values from saved logs
• get_all_secret_data.ipynb - get all features for all secret data
• get_secret_results.ipynb - use output of lstm_secret_data.ipynb to generate submission file
• merge_lstm_results.ipynb - use ensemble method on 5 LSTM models from lstm_secret_data.ipynb

models_utils

• Datasets.py - contains all datasets for PyTorch models
• GLOBALS.py - contains all global variables used in experiemnts
• utils.py - contains util functions

NN

• NeuralNetwork.py - NN model
• nn_utils.py - util functions for NN model

RF_XGB

• RandomForest.py - Random Forest model
• XGBoost.py - XGBoost model

Dataset

Link to Kaggle Dataset

Structure

The dataset is composed of 2 types of files:

• Acceleration data from Smartwatch sensor
• X,Y,Z data from Vicon

Results - Link to Leaderboard