This paper considers the need for generalizable bias mitigation techniques in machine learning due to the growing concerns of fairness and discrimination in data-driven decision-making procedures across a range of industries. While many existing methods for mitigating bias in machine learning have succeeded in specific cases, they often lack generalizability and cannot be easily applied to different data types or models. Additionally, the trade-off between accuracy and fairness remains a fundamental tension in the field. To address these issues, we propose a bias mitigation method based on multi-task learning, utilizing the concept of Monte-Carlo dropout and Pareto optimality from multi-objective optimization. This method optimizes accuracy and fairness while improving the model's explainability without using sensitive information. We test this method on three datasets from different domains and show how it can deliver the most desired trade-off between model fairness and performance. This allows for tuning in specific domains where one metric may be more important than another. With the framework we introduce in this paper, we aim to enhance the fairness-performance trade-off and offer a solution to bias mitigation methods' generalizability issues in machine learning.