Electrochemical biosensors promise real-time, high-quality monitoring of target compound levels in biofluid samples at the point-of-care. Each biosensor consists of a chip with multiple electrodes with different coatings to monitor several compounds simultaneously. However, the sensitivity and selectivity of biosensors decrease dramatically after each usage due to biofouling, a phenomenon caused by chemicals adhering to electrode surfaces. To improve detection accuracy and biosensor lifetime, we propose to characterize biofouling effects and correct fouled signals. However, analyzing biosensing signals is challenging due to the inherent patient-to-patient variability, complex electrode-to-electrode correlation, and limited knowledge of the underlying electrochemical processes. To overcome these difficulties under a repeated measurements setting, we characterize the signals for each patient as a functional mixed model whose coefficients are utilized as features that represent patient and electrode information. Therefore, changes between consecutive measurements can be detected by monitoring extracted features. We propose a series of nonparametric methods to predict and correct the biofouling-induced feature changes based on the feature similarity among patients. A case study illustrates the capability of the proposed method to predict and adjust fouled signals for new patients. The results suggest that the signal correction improves the detection accuracy of fouled biosensors.