PSI - Issue 70

Santhosh Kumar N V et al. / Procedia Structural Integrity 70 (2025) 440–446

441

1. Introduction In the construction industry, heuristics, trial-and-error as well as manual and approximate measurements were used to estimate the properties of building components like the compressive strength of lime mortar. Assessing compressive strength is among the most crucial of structural integrity evaluation and durability estimation of constructions due to impact they have on performance at practical applications. In addition, most approaches are excessively slow, costly, and require a high degree of effort, scope and precision for classifying structural building materials, especially composite materials, like lime mortar. In these situations, ML could assist greatly in negatively impacting cost and saving time for the construction industry by providing accurate calculations which could, in turn, help in formulating precise plans and frameworks to meet the minimal requirements with less tested performed. Lime mortar (LM) is a mixture of aggregates, water, and lime and has been widely used as a building material throughout history. LM has a number of advantages over traditional cement-based mortars in terms of workability and durability, and it is much more sustainable. There have been a number of studies in the past few years that have considered supplementary materials such as red mud or ground granulated blast furnace (GGBS) in order to improve the mechanical properties of LM and create a more sustainable product. Zhang et.al (2020) found red mud to be beneficial to the strength and durability of LM whereas Singh et, al. (2021) found GGBS had positive effects on the compressive strength of LM. One of the issues raised in these studies was a growing interest in the ability to use machine learning (ML) techniques to model the complex interactions of ingredients together and their mechanical properties to provide more confidence in predictions in future uses. Machine learning has already shown utility in a variety of applications beyond the materials realm. For example, machine learning has been used to make predictions dealing with material properties, formulations, and the search for new materials. Recent studies have explored the use of different ML algorithms like Random Forest, Support Vector Machines (SVM), and Artificial Neural Network (ANN) for predicting compressive strength of concrete and other construction materials. Chouhan et al. (2020) and Akinosho et al. (2021) investigated ML methods to predict compressive strength of high-performance concrete with reasonable success. The papers reflect the utility of ML in uncovering nonlinear relationships and interaction between input features, resulting in more accurate predictions and addressing some of the limitations of traditional regression-based methods that are not suited to measuring interactions in material properties. With regard to lime-based materials, there are several studies on the use of supplementary materials to modify the properties of lime mortar, but much of this research is based on traditional experimental methods. This clearly shows the need to include more data-centric studies (e.g., machine learning) to improve the prediction of compressive strength and reduce experimental proportions. Wang et al. (2019) and Kumar et al. (2020) both highlight the need for computational methods (e.g., ML) for construction material optimization, providing more sustainable and efficient construction material usage. This research's results are anticipated to add to the already substantial literature on machine learning in material science. Its purpose, by using a data-driven approach to modelling the compressive strength of lime mortar, has the aim of enhancing the formulations of the mortar, ensuring the cost effectiveness of the study, and minimising the construction sequence decisions. The ultimate goal of this study is to assist in the move towards effective, cost efficient, and sustainable construction practices, which are essential into the future. The paper is comprised of the following sections. The machine learning methods employed are Overviewed in Section 2 and entail: Random Forest (RF), Support Vector Machine (SVM), XGBoost, K-Nearest Neighbors (KNN) and Multiple Linear Regression (MLR). Section 3 includes a full description of the problem, some summary statistics of the input and output variables as well as a discussion of the model evaluation metrics. Section 4 provides the results and discussion concerning the model performance. Section 5 concludes the paper with a summary of the findings and closely-related suggestions for future work. 2. Machine learning methods In this study five supervised machine learning (ML) algorithms were utilized to predict the compressive strength of lime mortar, including Random Forest (RF), Support Vector Machine (SVM), XGBoost, K-Nearest Neighbors (KNN), and Multiple Linear Regression (MLR). The use of five methods has multiple advantages as Random