At Santander our mission is to help people and businesses prosper. We are always looking for ways to help our customers understand their financial health and identify which products and services might help them achieve their monetary goals.
Our data science team is continually challenging our machine learning algorithms, working with the global data science community to make sure we can more accurately identify new ways to solve our most common challenge, binary classification problems such as: is a customer satisfied? Will a customer buy this product? Can a customer pay this loan?
In this challenge, we invite Kagglers to help us identify which customers will make a specific transaction in the future, irrespective of the amount of money transacted. The data provided for this competition has the same structure as the real data we have available to solve this problem.
Table of Contents of The Notebook
- Summary
- Load Libraries and Dataset
- Exploratory Data Analysis
- Data Preparation
- Modelling
- Kaggle Submission
- Results
In this Kaggle competition the Santander Bank wants to predict "which customers will make a specific transaction in the future, irrespective of the amount of money transacted." [ Link to the competition ]
I achieved a score of 0.80469.
It seems the data has been duplicated with "fake" observations and in order to get a score over 0.90 a "magic" feature needs to be created (e.g. see this notebook). My focus in this notebook is not to find this "magic" feature, but to go through the data science process. Nonetheless, I did some experiments and it was an interesting learning experience, and perhaps something to explore further in the future.
Let's start.
Gaussian Naive Bayes was the best performing model of the 3 (Logistic Regression, Random Forest Classifier and Gaussian Naive Bayes), but also the fastest. RFC was the slowest one.
The main metric used during the modelling was the ROC-AUC. GNB also returned good results in other metrics such as Recall and Precission, where these are very important in an imbalanced dataset (which one is more importan will depend on specific cases, uses and policies).
After re-training the model with all the data and submitting the test data to Kaggle it returned a 80.47% score, which is the highest I ever achieved for this competition.
As next steps, a deeper look into the outliers would be a good idea. I tried to use a 3 times the Standard Deviation approach to identify outliers but it was deleting the whole dataset... Also, I didn't perform any data reduction since the models could handle all the dataset. However, experimenting with this would also be a good next step.
Due to limited size of GitHub dataset upload ability here I link to dataset from my Kaggle Santander Customer Transaction Prediction Dataset