Workflow, Markus John

Step 1: Load Data

• Loads the UCI Bank Marketing dataset

• No missing values

• The target variable 'y' indicates whether a client subscribed to a term deposit

• Target variable imbalanced (yes/no, ~11/89% class imbalance)

Step 2: Constant Values

• No constant values detected

Step 3: Correlations

• Rank Correlation was used to identify highly correlated variables

• euribor3m and nr.employed were removed due to high correlation

Logistic Regression:

• Categorical variables were transformed into dummy variables using One-to-Man

• Numerical variables were standardized using Z-score normalization

• Data was prepared for machine learning models

Logistic Regression Step 2:

• Used to predict whether a client subscribes to a term deposit

• The model estimates the probability that a client subscribes to a term deposit

• Coefficients were transformed into odds ratios using the Expression node to improve interpretability

• The analysis revealed that the variable duration had an exceptionally large influence on the model compared to the other predictors

Logistic Regression without duration:

• The variable duration was removed from the dataset

• Duration is only known after the phone call has taken place and is therefore not suitable for a realistic pre-call prediction scenario

• The influence of duration was identified during the odds ratio analysis

• From this point onward, all subsequent models are trained and evaluated without duration
  
  
  

Random Forest:

• Tree-based ensemble method as alternative to Logistic Regression

• No normalization needed

• Results evaluated on unseen test data

Boosted Trees (XGBOOST):

• Sequential tree-based ensemble method as alternative to Random Forest

• No normalization needed

• Results evaluated on unseen test data

Decision Tree:

• SMOTE applied only to training data to address class imbalance

• Chosen because results are easy to interpret and explain

Interactive Dashboard

• Bar chart shows how categorical variables are distributed across yes and no

• Box plot lets you select any numerical variable and see its spread

• Density plot shows the distribution of a selectable numerical variable

• Useful to spot differences between customers who subscribed and those who did not