FF_notes.txt

Interesting things to do with the data


Data Categories

Continuous: GPA, Grit, Material Hardship
Binary: Eviction, Layoff, Job Training (for a caregiver)


-Grit

-Job Training

-Combine data science and social science
-Use a variety of classifiers and thoughtfully explain each application
(Experimental idea, use a variety of classifiers and average the
results from some of them?)

-Selecting all binary or all continuous, and looking at the relations between accuracies for each
for a model

-Look at features and find relevant ones

-Focus on constructed features
-Maybe look for unconstructed feature that could be as useful?

-Single Feature choice (GRIT)


-Test Grit vs depression/mental happiness
-Drop mixing values for NA in train/test csvs





GENERAL APPROACH

1)
Write about the general problem
Describe a specific research question we will try to answer (How does the relevance of 
mothers and fathers change over time?)
Describe why it's relevant and important

2) Data processing decision making and rationale
hypothesize about relevant features?
List a bunch of classifiers that can be used to predict the things
Describe how certain classifiers will be of particular use

3) Evaluation and spotlight classifier

4) Analyze results
ROC curves/standard analysis metrics
Look at the strongest and weakest predictors of grit
Look at a feature we chose that confuses our model the most with regards to grit
(maybe threshold it with the average value of the numerical error?)
Look into things like why grit may be more complicated as a predictive/general feature
Correlate grit to some kind of happiness score?

5) Draw general statistical conclusions
Further directions of research
Theoretical approaches and applications to social work


-Feature selection ratio?
-mother, father, kid, teacher,

-Father later years
-Mother earlier years

-Comparison of how relevant mothers and fathers over time