An optimized machine learning-based unit for smart arsenic electrochemical biosensors

Biosensing platforms are analytical devices that incorporate biological receptors to detect and quantify specific analytes with high precision. Despite their effectiveness, a major limitation arises from the gradual decline in the activity and stability of immobilized bioreceptors on the electrode surface over time. This degradation necessitates frequent replacement or recalibration, thereby increasing operational costs and hindering large-scale commercialization. In the present study, a three-electrode electrochemical biosensor modified with gold (Au) nanoparticles was developed for the selective detection of arsenite, the trivalent and highly toxic form of arsenic frequently found in contaminated water resources. To improve analytical performance and predictive accuracy, machine learning algorithms were integrated into the biosensor’s analytical framework. Electrochemical response data, solution pH, enzyme lifespan, and storage temperature were considered as input parameters for model development. To further enhance model efficiency, parameter optimization was performed using two metaheuristic algorithms: the firefly algorithm (FA) and the gravitational search algorithm (GSA). These were applied to tune the parameters of two predictive models: the adaptive neuro-fuzzy inference system (ANFIS) and the Random Forest (RF) model. Among the hybrid configurations tested, the FA-RF demonstrated superior performance, achieving a coefficient of determination (R²) of ۰.۹۳ and a mean squared error (MSE) of ۰.۰۰۷. These outcomes, derived from biosensor data collected over a ۴۵-day operational period following enzyme and Au nanoparticle immobilization, highlight the potential of metaheuristic optimization and machine learning in enhancing the reliability, sensitivity, and lifespan of intelligent biosensing systems.

Keywords:

Artificial intelligence , ANFIS , Random Forest , Firefly algorithm , gravitational search algorithm