Which model was best? Report the results. Which variables were important? Is the result what you would expect?
data mining
Description
DataRobot Machine
Learning Challenge!
For this bonus project, you will need a license for the
DataRobot application. You can purchase the license here:
https://www.datarobot.com/datarobot-academic-payment/
…and select from one
of the four options available for student accounts:
· 1 week (7 days) for $15
· 2 weeks (14 days) for $30
· 3 weeks (21 days) for $45
· 9 months (275 days) for $60
You will also need to select a dataset from: https://www.kaggle.com/datasets
and/or a current challenge from: https://www.kaggle.com/competitions
Note: you do not have to submit your assignment to the
online competitions in order to receive credit for the assignment (though, you
certainly are welcome to do so, and there are cash prizes available).
Your assignment is to create and explain a predictive model
using DataRobot to apply multiple Machine Learning techniques to your dataset.
You will deliver a word doc that describes:
·
The dataset and the prediction challenge you are
tackling from kaggle
o
Be sure to describe the data in reasonable
detail here.
o
If you are not using a defined challenge, make
sure your selection of target (dependent) variables makes sense in the context
of your data. Ask if you have
questions here!
·
The results of your first “naïve” model (without
transforming any data or adding any new data to your set, just running
Autopilot in DataRobot).
o
Which model was best? Report the results. Which
variables were important? Is the result what you would expect?
·
Attempt to create a “blender” by using multiple
models at once. Does this improve your results from the naïve model?
o
Explain
and report.
o
Are the results growing more clear or more
opaque?
·
Try “feature engineering” by
transforming/adjusting/combining data in your dataset. Does this improve your
results?
o
Report the best results you can achieve so far.
·
Full credit will only be given if you have
improved on the naïve model via blending and feature engineering. Include
screenshots/reports for verification.
