Overview:

In order to predict customer churn in a credit card portfolio at American Express, I would follow a systematic approach which takes into account the available data, problem statement, business context, and the performance of different available models. The process would involve the following steps:

1. Define the problem statement:
   Firstly, I would clarify the problem statement and identify the specific objectives of the churn prediction model. This would include deciding on the definition of churn and understanding the business implications of accurate or inaccurate predictions.
2. Collect and prepare data:
   Next, I would gather the relevant data from different sources such as transaction records, customer demographics, credit scores, payment behavior, and customer service history. This data would then be cleaned, processed, and transformed into a format suitable for machine learning algorithms.
3. Explore and visualize data:
   Before applying any model, I would explore the data to identify trends, patterns, and relationships between different variables. This would involve statistical analysis, data visualization, and hypothesis testing techniques to gain insights into the data.
4. Select relevant features:
   Based on the data exploration, I would select the relevant features or variables that are most likely to have an impact on churn prediction and remove the ones that have no discernible relationship with churn.
5. Apply and evaluate models:
   In this step, I would apply various machine learning models such as logistic regression, decision trees, random forest, gradient boosting, and neural networks to the data and evaluate their performance using metrics such as ROC curve, confusion matrix, precision and recall. I would also tune the hyperparameters to optimize the performance of the model.
6. Validate the model:
   Once a suitable model is identified, I would validate its performance on a separate dataset to ensure that it is reliable and accurate in predicting churn.

Conclusion:

By following this process, I would be able to identify the most appropriate machine learning models to predict customer churn in a credit card portfolio at American Express. The selected model would be reliable, accurate, and aligned with the specific objectives of the churn prediction problem.

References:

• Customer Churn Prediction
• How to Approach a Machine Learning Project:
  Part 1 — Project Planning
• A Gentle Introduction to Exploratory Data Analysis