Google Professional Machine Learning Engineer
Get started today
Ultimate access to all questions.
You work for a telecommunications company, and you're tasked with building a machine learning model to predict which customers may fail to pay their next phone bill. The goal of this model is to proactively identify at-risk customers and offer them assistance such as service discounts and bill deadline extensions to mitigate the risk of non-payment. The dataset for model training is stored in BigQuery and includes the following predictive features: - Customer_id - Age - Salary (measured in local currency) - Sex - Average bill value (measured in local currency) - Number of phone calls in the last month (integer) - Average duration of phone calls (measured in minutes). Your mandate includes ensuring the model does not exhibit bias against disadvantaged groups while maintaining high accuracy. What approach should you take to achieve this?
You work for a telecommunications company, and you're tasked with building a machine learning model to predict which customers may fail to pay their next phone bill. The goal of this model is to proactively identify at-risk customers and offer them assistance such as service discounts and bill deadline extensions to mitigate the risk of non-payment. The dataset for model training is stored in BigQuery and includes the following predictive features: - Customer_id - Age - Salary (measured in local currency) - Sex - Average bill value (measured in local currency) - Number of phone calls in the last month (integer) - Average duration of phone calls (measured in minutes). Your mandate includes ensuring the model does not exhibit bias against disadvantaged groups while maintaining high accuracy. What approach should you take to achieve this?
Explanation:
Option D is the correct answer because it encompasses a comprehensive approach to ensuring fairness in the model. Defining a fairness metric that represents the accuracy across sensitive features allows for quantifying potential bias. Training the model with all features initially helps establish a baseline performance. Making predictions on a held-out test set ensures that evaluation is based on unseen data, which helps avoid overfitting. Reintroducing sensitive features after making predictions allows the calculation of fairness metrics for different customer groups, thereby offering a clear insight into potential bias. This method also supports iterative refinement, where further mitigation strategies can be applied based on the results of the fairness metric.