# Human activity recognition from smartphone sensors I studied whether smartphone accelerometer and gyroscope signals could classify twelve activities and postural transitions. I worked with a course version of a dataset from 30 volunteers aged 19–48, recorded at 50 Hz. I extracted time- and frequency-domain features from 128-sample epochs and compared four classifiers with ten-fold cross-validation. Multinomial logistic regression had the highest accuracy in the original notebook: 89.21%. ## Results | Classifier | Best epoch-level CV accuracy | | --- | ---: | | Multinomial logistic regression | 89.21% | | Scaled KNN | 87.57% | | LDA | 86.68% | | KNN | 74.45% | The table comes from [original_model_comparison.csv](results/original_model_comparison.csv). I retain the original notebook and [tuning results](results/original_cv_details.csv). These are the original coursework results; I have not rerun the full comparison with the current script.
Code and files The R script requires `tidyverse`, `caret`, `nnet`, and `e1071`. ```bash Rscript analysis.R ``` The script reads the included training and test recordings. It writes `results/rerun_model_comparison.csv` and `results/submission_final.csv` when run. The file list below shows the files currently committed. ```text Human-Activity-Recognition/ ├── .gitattributes ├── .gitignore ├── README.md ├── analysis.R ├── data/ │ ├── README.txt │ ├── RawData/ │ │ ├── Test/ │ │ │ ├── acc_exp03_user02.txt │ │ │ ├── acc_exp04_user02.txt │ │ │ ├── acc_exp07_user04.txt │ │ │ ├── acc_exp08_user04.txt │ │ │ ├── acc_exp17_user09.txt │ │ │ ├── acc_exp18_user09.txt │ │ │ ├── acc_exp19_user10.txt │ │ │ ├── acc_exp20_user10.txt │ │ │ ├── acc_exp21_user10.txt │ │ │ ├── acc_exp24_user12.txt │ │ │ ├── acc_exp25_user12.txt │ │ │ ├── acc_exp26_user13.txt │ │ │ ├── acc_exp27_user13.txt │ │ │ ├── acc_exp36_user18.txt │ │ │ ├── acc_exp37_user18.txt │ │ │ ├── acc_exp40_user20.txt │ │ │ ├── acc_exp41_user20.txt │ │ │ ├── acc_exp48_user24.txt │ │ │ ├── acc_exp49_user24.txt │ │ │ ├── gyro_exp03_user02.txt │ │ │ ├── gyro_exp04_user02.txt │ │ │ ├── gyro_exp07_user04.txt │ │ │ ├── gyro_exp08_user04.txt │ │ │ ├── gyro_exp17_user09.txt │ │ │ ├── gyro_exp18_user09.txt │ │ │ ├── gyro_exp19_user10.txt │ │ │ ├── gyro_exp20_user10.txt │ │ │ ├── gyro_exp21_user10.txt │ │ │ ├── gyro_exp24_user12.txt │ │ │ ├── gyro_exp25_user12.txt │ │ │ ├── gyro_exp26_user13.txt │ │ │ ├── gyro_exp27_user13.txt │ │ │ ├── gyro_exp36_user18.txt │ │ │ ├── gyro_exp37_user18.txt │ │ │ ├── gyro_exp40_user20.txt │ │ │ ├── gyro_exp41_user20.txt │ │ │ ├── gyro_exp48_user24.txt │ │ │ └── gyro_exp49_user24.txt │ │ └── Train/ │ │ ├── acc_exp01_user01.txt │ │ ├── acc_exp02_user01.txt │ │ ├── acc_exp05_user03.txt │ │ ├── acc_exp06_user03.txt │ │ ├── acc_exp09_user05.txt │ │ ├── acc_exp10_user05.txt │ │ ├── acc_exp11_user06.txt │ │ ├── acc_exp12_user06.txt │ │ ├── acc_exp13_user07.txt │ │ ├── acc_exp14_user07.txt │ │ ├── acc_exp15_user08.txt │ │ ├── acc_exp16_user08.txt │ │ ├── acc_exp22_user11.txt │ │ ├── acc_exp23_user11.txt │ │ ├── acc_exp28_user14.txt │ │ ├── acc_exp29_user14.txt │ │ ├── acc_exp30_user15.txt │ │ ├── acc_exp31_user15.txt │ │ ├── acc_exp32_user16.txt │ │ ├── acc_exp33_user16.txt │ │ ├── acc_exp34_user17.txt │ │ ├── acc_exp35_user17.txt │ │ ├── acc_exp38_user19.txt │ │ ├── acc_exp39_user19.txt │ │ ├── acc_exp42_user21.txt │ │ ├── acc_exp43_user21.txt │ │ ├── acc_exp44_user22.txt │ │ ├── acc_exp45_user22.txt │ │ ├── acc_exp46_user23.txt │ │ ├── acc_exp47_user23.txt │ │ ├── acc_exp50_user25.txt │ │ ├── acc_exp51_user25.txt │ │ ├── acc_exp52_user26.txt │ │ ├── acc_exp53_user26.txt │ │ ├── acc_exp54_user27.txt │ │ ├── acc_exp55_user27.txt │ │ ├── acc_exp56_user28.txt │ │ ├── acc_exp57_user28.txt │ │ ├── acc_exp58_user29.txt │ │ ├── acc_exp59_user29.txt │ │ ├── acc_exp60_user30.txt │ │ ├── acc_exp61_user30.txt │ │ ├── gyro_exp01_user01.txt │ │ ├── gyro_exp02_user01.txt │ │ ├── gyro_exp05_user03.txt │ │ ├── gyro_exp06_user03.txt │ │ ├── gyro_exp09_user05.txt │ │ ├── gyro_exp10_user05.txt │ │ ├── gyro_exp11_user06.txt │ │ ├── gyro_exp12_user06.txt │ │ ├── gyro_exp13_user07.txt │ │ ├── gyro_exp14_user07.txt │ │ ├── gyro_exp15_user08.txt │ │ ├── gyro_exp16_user08.txt │ │ ├── gyro_exp22_user11.txt │ │ ├── gyro_exp23_user11.txt │ │ ├── gyro_exp28_user14.txt │ │ ├── gyro_exp29_user14.txt │ │ ├── gyro_exp30_user15.txt │ │ ├── gyro_exp31_user15.txt │ │ ├── gyro_exp32_user16.txt │ │ ├── gyro_exp33_user16.txt │ │ ├── gyro_exp34_user17.txt │ │ ├── gyro_exp35_user17.txt │ │ ├── gyro_exp38_user19.txt │ │ ├── gyro_exp39_user19.txt │ │ ├── gyro_exp42_user21.txt │ │ ├── gyro_exp43_user21.txt │ │ ├── gyro_exp44_user22.txt │ │ ├── gyro_exp45_user22.txt │ │ ├── gyro_exp46_user23.txt │ │ ├── gyro_exp47_user23.txt │ │ ├── gyro_exp50_user25.txt │ │ ├── gyro_exp51_user25.txt │ │ ├── gyro_exp52_user26.txt │ │ ├── gyro_exp53_user26.txt │ │ ├── gyro_exp54_user27.txt │ │ ├── gyro_exp55_user27.txt │ │ ├── gyro_exp56_user28.txt │ │ ├── gyro_exp57_user28.txt │ │ ├── gyro_exp58_user29.txt │ │ ├── gyro_exp59_user29.txt │ │ ├── gyro_exp60_user30.txt │ │ ├── gyro_exp61_user30.txt │ │ └── labels_train.txt │ ├── activity_labels.txt │ └── example_submission.csv ├── notebook/ │ └── Human_Activity_Recognition_Notebook.ipynb └── results/ ├── original_cv_details.csv └── original_model_comparison.csv ```
## Limitations Epochs from the same person can enter different folds, so the reported accuracy does not estimate performance on unseen people. I filtered predictors before cross-validation, which may inflate the estimates. The withheld test labels prevent me from reporting test accuracy. ## Credits I worked on this project with **Yuxuan Xiang**. Project authors: **Laura Maria Fetz** and **Yuxuan Xiang**. We shared feature extraction and model selection; I wrote the notebook text and prepared its layout and figures. The dataset documentation in [data/README.txt](data/README.txt) credits Jorge L. Reyes-Ortiz, Davide Anguita, Luca Oneto, Xavier Parra, and collaborators.