#62: Scikit-learn 59:Supervised Learning 37: Project: Biometrics activity classifier

The video discusses a machine learning project to build a Biometrics activity classifier using real world data. Video is in three parts that talk about code to (a) get data, (2) preprocess data, (3) build model using Scikit-learn in Python.

Timeline
(Python 3.8)
[Note: The video lags audio in several places.]

00:00:00 - Outline of video
00:01:25 - Data source
00:02:05 - Story of sensors, biometrics/activity
00:03:40 - Data: folder system
00:04:56 - Data: files
------------------------------------------------
* * * PART - I * * *
------------------------------------------------
00:07:00 - Open Jupyter notebook
00:08:36 - Download data: set path
00:08:47 - Read data files: create .get_id()
00:11:23 - Read data files: create .clean_file_df()
00:26:50 - Read data files: put data from files into DataFrame
00:33:34 - Read data files: check DataFrame 'dfmain'
00:35:57 - Visualize data: Heatmap
00:40:40 - Visualize data: Interactive 3D scatter plot
------------------------------------------------
* * * PART - II * * *
------------------------------------------------
00:47:25 - Recap
00:48:16 - Slice dfmain: create DataFrame 'df'
00:50:53 - Get dummies: create DataFrame 'df_sen_dev'
00:52:42 - Create DataFrame 'dfr'
00:53:06 - Create new Features: initialize variables
00:53:55 - * * * CORRECTION * * *: 'Feature # 1' is a typo. Here, a list is created to be used later to create features
00:56:32 - Create new Features: Feature # 1
00:59:36 - Create new Features: Feature # 2
01:02:30 - Create new Features: Feature # 3
01:05:52 - Drop rows: create filter: based on values '!=0'
01:09:24 - * * * NOTE * * *: severe video lag from audio
01:11:20 - Drop rows: plot
01:11:56 - Drop rows: which rows were removed?
01:13:06 - Drop columns not needed
01:14:12 - Split 'dfmain' to 'X' and 'y'
01:16:05 - Remove features with zero variability
01:17:58 - Remove collinear features: create .remove_collinear_cols()
01:21:58 - Remove collinear features: Visualize
01:23:14 - Transform distributions: skew plot
01:24:00 - Transform distributions: create .q_transform()
01:25:28 - Transform distributions: update DataFrame
01:26:19 - Transform distributions: plot transformed data
01:26:36 - Standardize: using .StandardScaler()
01:27:47 - Join the sensor and device columns
01:29:19 - Check group count
------------------------------------------------
* * * PART - III * * *
------------------------------------------------
01:30:49 - Recap
01:31:13 - Split train and test
01:32:19 - Model: create .fit_predict()
01:34:19 - Model: create list of classifiers
01:40:47 - Model: for-loop to classify
01:44:14 - Model: results for accuracy and confusion matrix
01:49:32 - Model: re-run with all activity classes
01:58:59 - Model: re-run with 5 classes and drop a feature
02:01:16 - * * * CORRECTION * * * misspoke there! meant to say 'climbing stairs'
02:03:33 - Ending notes

#################
# Download data
#################
Citation-1: Gary M. Weiss, Kenichi Yoneda, and Thaier Hayajneh. Smartphone and Smartwatch-Based Biometrics Using Activities of Daily Living. IEEE Access, 7:133190-133202, Sept. 2019.

#################
# Code
#################