PSI - Issue 70

Abdul Musavvir et al. / Procedia Structural Integrity 70 (2025) 432–439

433

1. Introduction The HSC plays a significant role in the infrastructure development of the modern construction. Booming in urbanization causes tremendous population growth in urban regions. To accommodate this population, rise, tall buildings are constructed for commercial and residential in order to avoid land expansion in rural region. The construction of tall buildings increases the static and dynamic loads (Dead load, Imposed load, Wind load, and Earthquake load) on a building. So, the High Strength Concrete (HSC) is used to withstand these loads. The HSC is also used in more structures like bridges, railway decks, airstrips and docks which often experience heavy dynamic loading. The increasing usage of Cement rises environmental concerns of carbon footprint. Many researches are going on in replacing the cement content in concrete by utilizing the supplementary cementitious material from industrial wastes which poses cementitious properties (Prashant M. Dhamanage, 2020). The optimization of the mixes using these industrial waste becomes a complex calculation which needs to be accompanied with experimentation for better results. This traditional method delays the initial progress of the construction, increases the cost and the wastage due to experimentation. The Mahine Learning (ML) is used for developing prediction model all around the world by various fields of science and technology. The ML have opened new possibilities for predicting the properties of concrete with high precision (Torkan Shafighfard, 2024) (Priyanka Singh, 2023) (Priyanka Singh G. C., 2022). ML algorithms can analyze large datasets, identify intricate patterns, and model nonlinear relationships between input parameters and output responses. There is large dataset available for HSC with supplementary cementitious material from industrial wastes from many researches which can be applied to the ML to develop a predictive model for HSC. By leveraging historical data on concrete mix proportions and their corresponding strength values, ML models can learn from past observations and make highly accurate predictions for new mix designs (Manan Davawala, 2023). Numerous researches are conducted on developing a predictive model to predict properties of concrete using different ML algorithms like linear regression (Singh, 2024), decision tree regression, support vector machines, random forest, bagging regression, Adaboost, catboost, Xgboost and neural networks (Er. Tushant Seth, 2023) (H N Muliauwan, 2020) (John F. Vargas, 2024) (Sana Shabir Khan, 2022) (Suhang Yang, 2024) (Alexey N. Beskopylny, 2022) (MELTEM ÖZTURAN, 2008). The Xgboost, random forest and bagging regression were performing the best among all other algorithms in most of the researches conducted. The accuracy of the models was more pronounced in the 6 – 31 MPa range due to the larger size of data points. The studies found that age is the most significant variable influencing bearing capacity, followed by total cementitious materials. Higher values of total cementitious materials increase the compressive strength significantly, while lower values of fine aggregate and coarse aggregates decrease it significantly. The developed predictive model using the ML predicted the 28 th day compressive strength of HSC ranging from 60MPa to 100MPa. The model was developed with four algorithms (decision tree, linear regression, random forest and Xgboost) to compare and choose the best fitting algorithm for the available dataset consisting 355 data with ten parameter (cement, fine aggregate, coarse aggregate, water, super plasticizer, fly ash (Shashikant Kumar, 2024), ground granulated blast furnace slag (Vimal Rathakrishnan, 2022), nano silica and silica fumes (Adebanjo, 2024)). The developed predictive model was then validated by real time experimentation with eight different mix combinations. 2. Machine Learning Machine Learning (ML) is a department of artificial intelligence (AI) that empowers computers to learn patterns from data and progress their execution on assignments without being unequivocally modified. It includes creating calculations that can distinguish designs, make predictions, and adjust based on data. ML can be categorized into three categories; supervised learning, where models are prepared on labeled information to form expectations; unsupervised learning, which distinguishes covered up designs in information without predefined names; and reinforced learning, where a specialist learns ideal decision-making through trial and blunder in an environment. An ML algorithm is a set of mathematical and computational rules that a model follows to learn patterns from data and make predictions or decisions. These algorithms process input data, extract meaningful features, and adjust their internal parameters to improve accuracy over time. The choice of algorithm depends on factors such as data size, complexity, and the specific problem being solved.