Categorical Variables

Loan Purpose

Mosaic (Martial Status X Credit Risk X Gender)

Continuous Variables

PCA

χ-squared

Supervised Learning Results

Insights & Further Considerations

• The majority of customers in this data set are male, single, own a house, and have do ‘skilled’ work.
• The data is not representative for female customers
• Loan Purpose, Gender, Marital Status, and Housing had significant relationships with Credit Risk. The relationship between Job & Credit Risk was not significiant.
• 88.8% of variance is explained in PC1 through PC5. However, removing ‘Years’ (which is primarily in PC6) reduced the accuracy of every model by 1.0-2.5%
• The most accurate model was RF(n = 850)*

What went well

• I was able to start learning the basics of building interactive plots (plotly) and dashboards (flex dashboard) in R, and sharing R files
• The R community & open-source nature of the language was incredibly helpful

What didn’t

• Needed to spend more time in the ‘80%’ of data analysis
• Naming scheme interfered with dashboard render
• R provided an all-in-one package, but executing the overall dashboard was difficult. Other intricacies were problematic such as libraries & functions
• Building Marimekko/Mosaic Plots in plotly & ggplot
• Program stability

Future plans

1. Define a more clear-cut goal
2. Feature engineering: Dummy coding/one hot encoding categorical variables, creating categories for Age, Checking, & Savings, and combining
3. Fix current code
4. Testing the model with additional data
5. Create dashboard slides that work together