This notebook goes through various classification techniques with the use of the bank marketing dataset.
The following research questions were identified from the dataset, and have been answered throughout the analysis using eda, and machine learning methods;
- Do demographic factors such as the level of education, marital status, and type of job affect the success of the marketing campaign?
- Does a customer’s previous banking behaviour such as taking out a loan, or defaulting impact their decision on whether to subscribe to the term deposit or not?
- Does the time period in which the customer is contacted increase the effectiveness of the campaign?
- Which features are the most important when determining the effectiveness of the marketing campaign?
- Can the outcome of the marketing campaign be predicted using classification models? Which model performs the best?
The models have been optimized using hyper-parameter tuning and Recursive Feature Elimination (RFE with cross validation) for feature selection.
The following evaluation metrics were observed;
| Algorithm | ROC-AUC |
|---|---|
| Logistic Regression | 0.76 |
| CART | 0.70 |
| Random Forest | 0.77 |
| XGBoost | 0.76 |
It was concluded that the Random Forest classifier performs the best whilst the XGBoost and logistic regression models also having almost the same performance.