Predictive comparison and evaluation of ANNs and ANFIS as effective tools for modeling cellulase production by Bacillus mojavensis ND72

Abstract

Biological processes have traditionally been modeled using statistical and mathematical methods. These methods are often time-consuming and inefficient. The results obtained from these modeling techniques may not accurately model and predict outcomes in many processes. With the advancement of technology, artificial intelligence methods, especially artificial neural networks and adaptive structures such as ANFIS, have become powerful tools for such applications. This study employed both ANN and ANFIS models, each trained on 70% of the experimental data. The remaining 15% of the data, comprising combinations of cellulose, pH, temperature, and time along with their corresponding cellulase activity obtained from the conventional system, served as the testing set. The performance of both models was evaluated based on Mean Squared Error (MSE). The ANN model exhibited training and testing MSE values of 0.3687 and 0.8836, respectively, while the ANFIS model demonstrated significantly lower MSE values of 0.0009 and 0.0012, respectively. These low MSE values indicate acceptable levels of error, considering the limited size of the experimental dataset. Furthermore, the correlation coefficient (R) was calculated to assess the accuracy of both models. The ANN model exhibited an R-value of 0.9989, while the ANFIS model demonstrated a slightly higher R-value of 0.9991, indicating a strong correlation between the predicted and actual cellulase activity. The results demonstrate that both ANN and ANFIS models effectively predicted cellulase activity. However, the ANFIS model consistently exhibited superior performance, demonstrating closer agreement with the actual experimental values.

Graphical abstract