Optimization of biocementation responses by artificial neural network and random forest in comparison to response surface methodology

Enhancing Predictions of Acetate and Ethanol Production from Microbial Electrosynthesis Using Optimized Machine Learning Models

“…The XGBoost model, after hyperparameter optimization, achieved R 2 values of 1 for the training set and 0.877 for the testing set (Figure e). Moreover, Figure b,d,f displays residual plots that are near zero, with prediction error plots aligning approximately with a straight line (Figure a,c,e), indicating the conformity of the model’s prediction errors to a normal distribution …”

“…Models were then instantiated and trained using the “fit” function on the training set, followed by the application of the “predict” function to generate predictions on the testing set. In this study, we evaluated model performance using commonly used metrics in previous literature, namely, mean squared error (MSE) and coefficient of determination ( R 2 ). ,, MSE and R 2 were calculated using the following equations M S E = 1 N ∑ i = 1 N ( d i − g i ) 2 R 2 = 1 − ∑ i = 1 N ( d i − g i ) ∑ i = 1 N ( d i − d m ̅ ) where d i represents the predicted value by the model, g i denotes the actual value, N is the total number of data points, and d m is the mean of the actual values for the dependent variable.…”

“…Moreover, Figure 4b,d the conformity of the model's prediction errors to a normal distribution. 28 Upon comparing the three optimized models, it became apparent that their accuracies in predicting the acetate concentration based on the features time, current, applied potential, temperature, pH, and IC concentration were largely comparable. However, XGBoost outperformed the other models with higher R 2 and lower MSE values.…”