Predictive Intelligence for Compressive Strength Estimation in Cement Industry Using ML

Abstract

In the global construction industry, concrete remains a fundamental material, with over 10 billion tons 
produced annually, contributing to 8% of global CO₂ emissions. The compressive strength of concrete 
is a key quality indicator, directly affecting the durability and safety of over 70% of civil infrastructures. 
Despite its importance, many cement manufacturing industries still rely on manual mix ratio estimations 
and laboratory testing, which can take up to 28 days, leading to project delays, cost overruns, and 
resource wastage. Additionally, variability in raw materials, human error in proportioning, and lack of 
real-time decision-making hinder the optimization of mix design. To address these issues, a machine 
learning-driven framework is proposed for predicting concrete compressive strength using data 
analytics. The system begins with dataset collection, incorporating variables such as cement, water, fine 
aggregate, coarse aggregate, fly ash, and superplasticizer content, along with age. The dataset undergoes 
preprocessing to handle missing values, normalize input features, and split data for training and testing. 
Initially, the K-Nearest Neighbors (KNN) regressor is implemented as a baseline model, which shows 
moderate accuracy but struggles with high-dimensional correlations. The proposed model employs an 
Extra Trees Regressor, an ensemble learning technique that uses multiple randomized decision trees for 
robust prediction. This model enhances feature importance analysis and minimizes overfitting while 
significantly improving prediction accuracy and reducing error metrics such as RMSE and MAE. The 
proposed solution offers a faster, scalable, and more precise alternative to traditional strength prediction 
methods, paving the way for real-time quality control in smart cement manufacturing systems.