Enhanced Compressive Strength Prediction Using Concrete Mix Parameters and Feature Importance Analysis: Using Al-ternative ML Algorithms


Çakmak T., Ustabaş İ.

4th International Civil Engineering & Architecture Conference (ICEARC'25), Trabzon, Türkiye, 17 - 19 Mayıs 2025, ss.1-8, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: Trabzon
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.1-8
  • Recep Tayyip Erdoğan Üniversitesi Adresli: Evet

Özet

Concrete is one of the best commonly applied building materials due to its various advantages as such high strength, durability, resistance to high temperatures, etc. However, due to the CO2 emissions abput the production and consumption of the raw material cement, research has long been conducted into various alternatives. Various supplementary cementitious materials with pozzolanic properties are at the leading of such options. In addition to reducing the CO2 emissions of these materials, it is possible to improve various properties, especially strength and durability. However, obtaining the properties that occur when various materials are used together with concrete results from long laboratory processes and labour. Therefore, ML algorithms allow us to reduce this time and effort. In this study, a variety of ML algorithms such as RF, ET and GBR were used to predict the properties of concretes with different mix parameters such as different proportions of FA, GGBS, aggregate quantity and curing time. In addition, feature importance analyses based on different algorithms were performed to elucidate the background functioning of these algorithms. As a result of the analysis, the GPR algorithm showed the highest prediction and generalisation performance with an R2 value of 0.871. The ET and RF algorithms follow this performance. In addition, as a result of the feature importance analyses, the most important mixing parameters for all algorithms were found to be cement ratio and setting time. The values obtained as a result of the study show the suitability of using ML algorithms to detect different properties of concretes.