Machine Learning based Predictive Modelling for Sustainable Product Design

Silica Fume (FC) is an eco-friendly concrete which is commonly used in construction industry as a building material. With the growing awareness of environmental issues, developing predictive Machine Learning (ML) models becomes most important for optimizing the use of SF in concrete production. This research aims to develop predictive models for Silica Fume Concrete (SFC) properties to enhance Sustainable Product Design (SPD) through ML techniques. Utilizing a comprehensive dataset of SFC, the research focuses on analysing the importance of input selection, isolating six key variables. The feature selection employs an Improved Binary Cuckoo Search Algorithm (IBCSA) enhanced with a Chi-square filter and chaotic map for optimal feature identification. Predictive modelling is done using Decision Trees (DTs), with ensemble methods like bagging and boosting to mitigate overfitting and improve accuracy. The experimental results are validated on SFC and the proposed ML based SPD achieved better accuracy rate at 98.49%, Mean Square Error (MSE) of 1.04, Mean Average Error (MAE) of 0.73 when compared to the results of the existing ML based predictive models such as Gene Expression Programming (GEP) and DT Regression- MultiLayer Perceptron (DTR-MLP).