Enhancing the electrochemical detection of chloride using a machine learning approach

Electrochemical sensors are widely used in many applications, offering a competitive advantage over traditional lab-based methods because of their portability, price point, and sensitivity. However, there are also challenges related to the use of electrochemical sensors, such as electrode fouling or undesirable signal-to-noise ratios. Machine learning (ML) offers promising means to address these challenges. Herein, we develop an ML model to improve the efficiency and reusability of a point-of-use electrochemical sensor for chronopotentiometric chloride ion detection. We evaluated five ML models to predict chloride ion concentration, ultimately selecting Random Forest (RF) as the optimal model architecture based on model performance comparisons on the test set (test MAE of 2.9 mM and test R 2 score of 0.99). We accurately predict a wide range of chloride ion concentrations (15 to 250 mM) using a single current density (600 A/m 2 ), improving the operational efficiency of the sensor. Additionally, chloride ion concentrations could be reliably predicted using just 0.1 seconds of data using ML instead of up to 5 seconds without ML, reducing both the amount of time needed to quantify chloride and the opportunity for sensor fouling. We show that the reduced quantification time due to the RF model with the addition of a feature to correct for sensor signal drift enabled the single-use sensor to be reused at least 20 times, thereby extending the sensor lifetime. This work highlights the use of ML to advance the analytical performance and lifetime of electrochemical sensors. • A Random Forest machine learning (ML) model improves the performance of an electrochemical sensor for chloride ion detection in solution. • The model requires only 0.1 seconds of data, reducing the amount of time needed to detect chloride in solution by over 50-fold compared to the standard procedure. • The use of the model eliminates the need to sweep through multiple current densities and use multiple sensors in order to quantify an unknown chloride ion concentration. • The ML approach extends the number of times a single sensor can be reused up to at least twenty times.