Multilayer perceptron method of artificial neural network in classifying concrete compressive strength

KỶ YẾU HỘI THẢO KHOA HỌC QUỐC GIA CITA 2020 “CNTT VÀ ỨNG DỤNG TRONG CÁC LĨNH VỰC” Multilayer Perceptron Method of Artificial Neural Network in Classifying Concrete Compressive Strength Thi Phuong Trang Pham Danang of Technology and Education, The University of Danang trangpham3112@gmail.com Abstract– Compressive strength is a basic feature of concrete, it reflects the bearing capacity of concrete. Therefore classifying concrete compressive strength (CCS) plays a vital role. When CCS classification accuracy is improved which will be grounded to calculating bearing capacity, deformation of concrete and reinforced concrete structures better. The objective of this paper is to use multilayer perceptron (MLP) model for classifying CCS. The predictive accuracy of model was compared with several other model including support vector machine (SVM), Navie Bayes (NB) and decision tree (DT). Analytical results showed the MLP model was superior to other comparative models for concrete dataset. Particularly, the MLP was the best model achieving the highest results (92.524% of accuracy). Therefore, MLP model is considered a suitable tool to classify CCS dataset. Keywords: Concrete compressive strength, multilayer perceptron, support vector machine, Navie Bayes, Decision Tree. 1 Introduction In the field of construction nowaday, one of the common building materials that play a very important role in the construction of infrastructure of all buildings is concrete. Concrete is widely used in the construction of structures, foundations, pavement surfaces, bridges, roads, runways, structures in parking lots, dams The concrete compressive strength (CCS) can be considered on of characteristics of concrete. Therefore, there are diversity of researches study about CCS. For example, Prayog and et al applied artificial intelligence models including Evolutionary Support Vector Machine Inference Model (ESIM) and K-means Chaos Genetic Algorithm (KCGA) for optimizing performance concrete mix design [1], Gideon and et al studied compressive strength of concrete produced from different brands of Portland cement [2] or Metwally proposed model which was studied for different concrete mixes to predict compressive strength of Portlan cement concrete with age [3]. One of the problems of classification is how to use appropriate methods to fit the model depending on the nature of data. Nowaday, there are more and more algorithms applied to hanlde classification problem. Artificial intelligence (AI) is currently developing at a fast pace and has made important contributions to production, business, services and human life. AI methods can be considered as efficient tools for solving real-world engineering problems. Many AI techniques have been applied in construction engineering and construction management [4, 5] and they are usually used to handle prediction and classification problems. Moreover, neural networks are the popular classification tools used in classification problem. Multilayer perceptron (MLP) method of artificial neural network (ANN) is one of the foundational algorithms that helps neural networks work. This algorithm was commonly used in the 1980s in machine learning. MLP includes a series of fully interconnected layers of nodes where there are only connections between adjacent layers [6]. MLP algorithm also was applied in various research in classification. For example, Dash et al used MLP mothod for real‐world data classification under uncertainty [7] or Ncibi et al applied MLP based a preprocessing and hybrid optimization task for data mining and classification [6]. Besides, Kulala et al improved the performance of MLP model in classification by several optimization algorithms [8]. Therefore, in this study the author used MLP to classify CCS dataset. To evaluate the effectiveness of proposed model, this model is compared with other popular AI models including support vector machine 82 Thi Phuong Trang Pham (SVM), Naïve Bayes (NB) and decision tree (DT). The predictive performances of each model are compared in terms of accuracy, precision and sensitivity. The rest of this paper is organized as follows. Section 2 elucidates MLP model, and performance evaluation methods. CCS dataset and preparation dataset is presented in Section 3. Section 4 shows analytical results and finally, the conclusions and future study directions were drawn in Section 5. 2 Methodology 2.1 Artificial Neural Network (ANN) The concept of ANN is basically introduced from the subject of biology where neural network plays a important and main role in people’s body [9]. The ANN consists of 3 main components, they are the input layer and the output layer consist of only 1 layer, the hidden layer can have 1 or more layers depending on the specific problem. 2.2 Multilayer perceptron (MLP) model MLP is one of the most commonly used neural network architecture. A MLP is the implemnet of multiple layers of perceptrons connected side by side, forming a simple nerve feedforward controller. The feedforward network is a network of one or more neuron layers, in which the signal conductors only travel in one direction from input through layers, to output, hence a MLP usually consists of three or more layers: an input layer, one or more hidden layers and an output layer [10]. This MLP is useful in non-linear functions that a single perceptron cannot implement. Fig.1 showed the architectural graph of MLP. Input layer Hiden layer Output layer Fig.1. Architecture of multilayer perceptron model. 2.3 Evaluated measurement The performance measures that were used to assess the prediction of the proposed system including the accuracy, precision and sensitivity. Whereas the accuracy is the most important issue to estimate the results. Accuracy can be defined as the degree of uncertainty in a measurement with respect to an absolute standard. The predictive accuracy of a classification algorithm is calculated as follows, 83 KỶ YẾU HỘI THẢO KHOA HỌC QUỐC GIA CITA 2020 “CNTT VÀ ỨNG DỤNG TRONG CÁC LĨNH VỰC” tp tn Accuracy tp fp tn fn + = + + + (1) Whereas tp – true positive values is the number of correctly recognized class example, tn – true negative values is the number of correctly recognized examples that do not belong to the class representing accurate classifications, fp – false positive value is the number of examples that are either incorrectly assigned to a class or fn – false negative value is number of examples that are not assigned to a class refers to erroneous classifications. Precision and sensitivity are two extended versions of accuracy. Precision measures the reproducibility of a measurement, whereas sensitivity – also called recall – measures the completeness. Precision in Eq. (2) is defined as the number of true positives as a proportion of the total number of true positives and false positives that are provided by the classifier. Sensitivity in Eq. (3) is the number of correctly classified positive examples divided by the number of positive examples in the data. In identifying positive labels, sensitivity is useful for estimating the effectiveness of a classifier. tp Precision tp fp = + (2) tp Sensitivity tp fn = + (3) 3 Concrete compressive strength (CCS) dataset The CCS dataset was obtained from a University of California, Irvine data repository (https://archive.ics.uci.edu/ml/datasets/concrete+compressive+strength), including a total of 8 features – cement, blast furnace slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate and age. The input attibutes information of CCS dataset is showed in table 1. The output of original dataset is the value of concrete compressive strength, the main goal of this paper is classification therefore the author will convert these value to nomal value. Following many research classification of performance concrete based the compressive strength of concrete has been suggested: (a) Ordinary Concrete: CCS in the range of 10 to 20 MPa. (b) Standard/Normal Concrete: CCS in the range of 25 to 55 MPa. (c) High-Performance Concrete : CCS in the range of 60 to 100 MPa. (d) Very High-Performance Concrete: CCS in the range of 100 to 150 MPa. (e) Exceptional Concrete: CCS more than 150 MPa. Thus, the ouput of new dataset is one of five ratings of the compressive strength of concrete. Table 1. Statistical input attributes of CCS dataset. Attributes Unit Minimum Maximum Average Standard deviation Cement Kg/m3 102.0 540.0 281.168 104.506 Blast Furnace Slag Kg/m3 11.0 359.4 107.277 61.884 Fly Ash Kg/m3 24.5 200.1 83.862 39.989 Water Kg/m3 121.8 247.0 181.567 24.354 Superplasticizer Kg/m3 1.7 32.2 8.486 4.037 84 Thi Phuong Trang Pham Coarse Aggregate Kg/m3 801.0 972.919 77.754 Fine Aggregate Kg/m3 594.0 992.6 773.580 80.176 Age Day 1.0 365.0 45.662 63.170 4 Analytical results Table 2 compares the performances of MLP model and other AI models using new CCS dataset. High values indicate favorable performance and vice versa. From table 2, MLP model showed the best performance with the highest values of accuracy, precision and sensitivity (92.524%, 91.237% and 93.014%, respectively). SVM model also yielded the quite high with 84.272% of accuracy, 84.110% of precision and 83.717% of sensitivity. From the table 2, we can see, NB and DT models had 82.913% and 84.446% of accuracy, respectively. Fig 1, 2 and 3 respectively performed accuracy, precision and sensitivity values of MLP and compared AI models. Table 2. Prediction performance comparison Model Performance measure Accuracy (%) Precision (%) Sensitivity (%) SVM 84.272 84.110 83.717 NB 82.913 82.242 81.916 DT 84.466 82.362 80.563 MLP 92.524 91.237 93.014 Fig. 2. Performance comparison of all models in term of the accuracy 84,272 82,913 84,466 92,524 75 80 85 90 95 SVM NB DT MLP Accuracy (%) 85 KỶ YẾU HỘI THẢO KHOA HỌC QUỐC GIA CITA 2020 “CNTT VÀ ỨNG DỤNG TRONG CÁC LĨNH VỰC” Fig. 3. Performance comparison of all models in term of the precision Fig. 4. Performance comparison of all models in term of the sensitivity 5 Conclusion In this paper, we proposed the MLP of ANN model to classifying CCS data. After that, the author compared MLP model with other AI models with the same dataset. Three performance measures, including accuracy, precision and sensitivity were utilized to compare the predictive performance of examined models. Overall, the MLP was the best model for the compressive strength of concrete. Partically, it achieved the highest values of accuracy, precision and sensitivity 92.524%, 91.237% and 93.014%, respectively. The contribution of this paper is to improve CCS prediction. It is clear that predicting correctly CCS, people can more accurately caculate other parameters of concrete in the future. In the following studies, the author hopes to optimize MLP model to enhance the accuracy of model which can apply more datasets. References 1. D. Prayogo, M.-Y. Cheng, D. Wibowo, Artificial Intelligence Approaches for Optimizing High- Performance Concrete Mix Design, 2012. 2. G. Bamigboye, A. Ede, C. Egwuatu, J. Jolayemi, O. Olowu, T. Odewumi, Assessment of Compressive Strength of Concrete Produced from Different Brands of Portland Cement, Civil and Environmental Research 7 (2015) 31-38. 84,11 82,242 82,362 91,237 75 80 85 90 95 SVM NB DT MLP Precision (%) 83,717 81,916 80,563 93,014 70 75 80 85 90 95 SVM NB DT MLP Sensitivity (%) 86 Thi Phuong Trang Pham 3. M.a.a. Abd elaty, Compressive strength prediction of Portland cement concrete with age using a new model, HBRC Journal 10(2) (2014) 145-155. 4. J.-S. Chou, C. Lin, Predicting Disputes in Public-Private Partnership Projects: Classification and Ensemble Models, Journal of Computing in Civil Engineering 27(1) (2013) 51-60. 5. M.-Y. Cheng, N.-D. Hoang, Risk Score Inference for Bridge Maintenance Project Using Evolutionary Fuzzy Least Squares Support Vector Machine, Journal of Computing in Civil Engineering 28(3) (2014) 04014003. 6. K. Ncibi, T. Sadraoui, M. Faycel, A. Djenina, A Multilayer Perceptron Artificial Neural Networks Based a Preprocessing and Hybrid Optimization Task for Data Mining and Classification, International Journal of Econometrics and Financial Management 5(1) (2017) 12-21. 7. T. Dash, H.S. Behera, A comprehensive study on evolutionary algorithm-based multilayer perceptron for real-world data classification under uncertainty, Expert Systems 36(1) (2019) e12327. 8. H. Kulala, K. Rani, Advancements in Multi-Layer Perceptron Training to Improve Classification Accuracy, International Journal on Recent and Innovation Trends in Computing and Communication 5 (2017) 353-357. 9. V. Sharma, S. Rai, A. Dev, A Comprehensive Study of Artificial Neural Networks, 2012. 10. A.T. Azar, S.A. El-Said, Probabilistic neural network for breast cancer classification, Neural Computing and Applications 23(6) (2013) 1737-1751. 87