IABC-MLP Model Analysis for Enhancing Concrete Strength Prediction

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Mar 22 2024

In a paper published in Scientific Reports, researchers introduced an advanced method for predicting concrete compressive strength, employing an improved artificial bee colony algorithm (IABC) combined with a multilayer perceptron (MLP) model.

Overcoming the limitations of conventional models, the enhanced algorithm addressed issues like local optima and slow convergence by incorporating a Gaussian mutation operator. Dubbed IABC-MLP, this model demonstrated superior accuracy in capturing the nonlinear relationship between compressive strength and influencing factors compared to traditional methods.

Comparative analyses showed that IABC-MLP outperformed ABC-MLP and particle swarm optimization (PSO) coupling algorithms in training accuracy. Additionally, the technique showcased superior performance when compared to individual learning algorithms such as MLP, decision tree (DT), support vector machine regression (SVR), and random forest algorithms (RF), indicating the effectiveness of heuristic algorithms in concrete compressive strength prediction.

Related Work

Past research on predicting concrete compressive strength has grappled with the complexities of concrete composition, curing conditions, and nonlinear relationships, often resorting to empirical regression models. While machine learning, particularly neural networks like MLP, has shown promise, overfitting and reliance on initial parameters persist. Challenges include the difficulty in capturing the intricate interactions among various factors influencing concrete strength and the need for adaptability in existing models to account for changing material properties over time.

Model Components and Enhancements

The model comprises an MLP neural network and an optimization algorithm, the primary ABC algorithm. MLP, featuring an input, hidden, and output layer, excels in nonlinear mapping, providing high fault tolerance and adaptability. However, challenges such as overfitting persist. Inspired by bee foraging behavior, the ABC algorithm conducts global and local searches in iterations.

The researchers proposed improvements to enhance convergence speed and accuracy, including introducing chaotic sequence initialization for honey sources and implementing a Gaussian mutation mechanism to enhance local search. These enhancements aim to optimize the search space and improve local exploration capability, addressing slow convergence and local optima trapping shortcomings.

Integrated Strength Prediction

The strength prediction model proposed in this study integrates an MLP neural network, renowned for its capability to handle complex nonlinear relationships, with an IABC algorithm. The IABC-MLP algorithm tackles challenges such as poor network stability and susceptibility to local optima, which MLP networks commonly encounter.

Within the IABC-MLP framework, the IABC algorithm optimizes the initial weights and thresholds of the MLP, resulting in enhanced global search capability and improved model accuracy and stability. Individual bees in the ABC algorithm represent solutions comprising weights and biases for the MLP, optimizing them through search and selection operations.

By minimizing prediction errors, the IABC-MLP model achieves more accurate predictions of concrete strength. Evaluation of model accuracy employs mean absolute error, root mean square error, and correlation coefficient as performance indicators, providing a comprehensive assessment of prediction accuracy. These metrics quantify the average error, sensitivity to outliers, and correlation between predicted and actual values, respectively, ensuring robust evaluation of the model's predictive capability.

Model Parameters and Optimization

The strength prediction model integrates an MLP neural network with an IABC algorithm, targeting issues like poor network stability and susceptibility to local optima commonly encountered in MLP networks. The IABC-MLP framework optimizes initial weights and thresholds using the IABC algorithm, enhancing global search capability and model accuracy. After selecting influential variables and preprocessing data, researchers fine-tune the model parameters.

The MLP model utilizes random initialization, rectified linear unit (ReLU) function for hidden layers, and linear function for the output layer. Cross-validation determines the optimal number of hidden layers and nodes, with two hidden layers and 50 and 25 nodes showing the most minor test errors.

The number of hidden layer nodes remains constant for PSO, ABC, and IABC algorithms. PSO employs 50 particles and 100 iterations with a learning rate and inertia factor set to 0.5 and 0.3, respectively. ABC also uses 50 bees and 100 iterations, improving the initial solution space and search mechanism. Parameter settings for the improved ABC are identical to the standard ABC algorithm.

IABC-MLP Superiority Analysis

The comparison analysis revealed that the IABC-MLP model outperforms single and coupled models in predicting concrete compressive strength. Compared to single models such as MLP, DT, SVR, and RF, the IABC-MLP model demonstrated superior performance across evaluation metrics like mean absolute error, root mean squared error, and correlation coefficient.

In contrast to the PSO-MLP and ABC-MLP coupled models, the IABC-MLP model showcased better fitting ability and convergence speed, attributed to its enhanced global optimization and adaptation capabilities. Overall, the IABC-MLP model is a promising approach for accurate and stable concrete strength prediction, offering valuable insights for future research in this domain.

Conclusion

To sum up, the study introduced the IABC-MLP model for predicting concrete compressive strength, achieving high accuracy and fast convergence. Comparative analyses highlighted its superiority over single and coupled models, demonstrating enhanced fitting quality and generalization ability. The model showed promise for real-world applications, yet further research was needed to extend its applicability to diverse environmental conditions.