stephanieleevillanueva / classification_and_d3

Predicting hypertension using classification algorithms and basic visualization of results using d3

classification_and_d3

Predicting hypertension using classification algorithms and basic visualization of results using d3

####Folder structure:

d3

contains .html files and .csv data files used to create the d3 bar graphs.

* accuracy.html html file with css and javascript. Uses ``d3.js`` (with tooltip) library to generate the graph * pivot_accuracy.csv dataset used by accuracy.html * recall.html html file with css and javascript. Uses ``d3.js`` (with tooltip) library to generate the graph * pivot_recall.csv dataset used by recall.html * feature_importance.html html file with css and javascript. Uses ``d3.js`` library and also uses transition to generate the graph

sql

contains .sql scripts to generate final datasets used in classification models.

* sql_tables.sql creates table schemas * final_tables.sql selects only columns needed from ``sql_tables.sql`` and creates new filtered ``.sql`` tables * script_raw.sql joins tables from ``final_tables.sql`` and generates a single ``.sql`` table with data in its original form to be used for analysis. * script_converted.sql takes ``script_raw.sql`` generated table and converts selected columns into binary form.

data

contains .dat files relating to hypertension, pulled from www.cdc.gov website for year 2011-2012. This dataset is used in classifying people at risk for hypertension.

py_and_ipynb_files

contains .py helper files for database connection using python and .ipynb notebooks for classification modeling and results analysis

* pass_.py stores user's password for database * postgresql_conn.py defines ``sqlalchemy`` engine connection parameters * cdc_data.ipynb populates database tables with data from the ``.dat`` files * cdc_analysis.ipynb splits dataset into training and test data to build classification models and measure model metrics such as ``accuracy``, ``precision``, ``f1`` and ``roc curve``. Also uses sklearn feature called ``rfe`` or recursive feature elimination to reduce number of dimensions used in building the models.