Classification algorithms have been widely used to solve data-driven fault diagnostics problems. The number of classification algorithms used has been increasing in recent years. Each classification algorithm has its own strengths and weaknesses, and the accuracy of classifiers changes with the different features used for training. As a result, traditional methods of selecting an appropriate classification algorithm, including domain expertise and trial and error, are becoming complex and difficult to employ. Classifier fusion has been used to solve this problem of selecting an appropriate diagnostic algorithm, and it also improves the generalizability of an algorithm. The performance of a classifier fusion algorithm is governed by the combination rule adopted for fusing multiple classifiers and how the bias and variance are balanced by the combination rule. However, research still needs to determine which combination rule optimally balances the bias and variance during classifier fusion. Therefore, this research develops a fusion methodology that combines the classifiers by balancing the bias and variance. This methodology reduces the number of false negatives and positives, thereby improving the overall accuracy of the algorithm for fault detection. A cost function that considers bias and variance errors was developed to evaluate the performance of the algorithm. Sequential quadratic programming-based optimization was employed to find the optimal combination of classifiers and balance of bias and variance. The developed algorithm was used for fault diagnosis of analog circuits, and the results indicate that the developed fusion approach improved diagnostic accuracy over existing classifier fusion techniques.