2. The Flow
Data Exploration
•Create charts and dashboard in tableau to do data exploration
Build Prediction
Model
• Dividing the dadaset into training and testing datasets.
•Training a model by using Decision Tree, Random Forest and Deep Belif Netwok
•Train the best model for each algorithm with all the data
•Save all the models to R objects and make ready for Azure
Azure Project
•Creating the model in Azure.
•Deploying the web serivce.
•Generateing the output based on the input data.
Web Applicaton
•creating the front-end using C# and asp.net
4. Model Training
How to deal with unbalanced datasets
It results in a highly biased mode ageist unsatisfied class
Solution:
Under-sampling
Cost Sensitive Learning
6. Model Training
The models
We eventually selected 6 variable
Trained using
Decision Tree
Random Forest
Deep Belief Network
Layer 1 Layer 2 Layer 3 Batch
Size
Learning
Rate
Accuracy Kappa
50 200 200 5 0.1 0.674629 0.350697
200 50 50 5 0.1 0.673806 0.349025
200 200 50 2 0.1 0.672982 0.347466
50 50 50 5 0.1 0.672982 0.347352
200 50 50 5 0.8 0.672982 0.347296
50 200 50 5 0.1 0.672982 0.347296
200 50 50 2 0.4 0.672158 0.345567
200 50 200 2 0.1 0.672158 0.34551
50 50 50 5 0.4 0.672158 0.34534
Model Name Accuracy Kappa Kappa (with all variables)
Random Forest 0.76 0.17 0.17
Decision Tree 0.40 0.04 0.06
DBN 0.61 0.04 0.09
7. The Azure Machine Learning
Web service input and model.zip are
connected to the same node to Execute R
Script.
Enter Data Manually is connected to another
node to Execute R Script, and they are all
input data.
Web service output is connected to the
output node of Execute R Script as the
output data.
8. Clients
1. Using ASP.NET and C#
Web Application with Visual Studio
Add API key and URL from Azure to
make connections
https://docs.microsoft.com/en-us/azure/app-service-web/media/web-sites-dotnet-get-started/create_app.png
11. Contribution Pie Chart
40.00%
40.00…
20.00…
Contribution
Amin Nishtha Fangning
Amin:
• Cleansing of data set
• Training and testing the data set
• Creating model objects for azure machine
learning
Nishtha/Alex:
• Data exploration using tableau
• Creating azure machine learning
• Creating web application using c# and
asp.net using visual studio and deploying the
web services
• Create report