The purpose of this notebook is do to a rudimental analysis of the Bank Marketing Data Set to reach a reasonable degree of accuracy when predicting whether a client will subscribe to a term deposit, and then perform a simple Bayesian search using the scikit-optimise Python library to find a (near) optimal learning rate.
The data is related with direct marketing campaigns (phone calls) of a Portuguese banking institution. The classification goal is to predict if the client will subscribe a term deposit (variable y). - UCI Machine Learning Repositry
In a business setting, a detailed analysis of this dataset might be done to determine how to be go about a telemarketing campaign. By performing an accurate prediction, we get a better sense of which features most affect the chance of a customer subscribing to a particular service provided by the bank. The bank can then focus their efforts on the features which increase the likelihood of the customer subscribing to a service.
From an academic standpoint, this analysis is a good way to understand how to use the scikit-optimise library to perform Bayesian Optimisation. This may not be the best library to perform a Bayesian search, however this is well known to be a easy to use and provide a wide range of functionality. From some prior reading, this dataset is imbalanced. A major challenge when analysing this will arise from how to handle this, without causing a large bias in the results or overfitting due to the majority label.
Below is the Bayesian search algorithm used. Specifics have been removed by can be found in the notebook. Click on the image below to go to a gist for this code.
