Concrete Strength Prediction Using MATLAB and Machine Learning

A student-laptop project that predicts the compressive strength of concrete from its mix proportions and age, and then uses the model in reverse to propose cheaper and lower-CO2 mixes. It is trained on real laboratory data - the UCI Concrete Compressive Strength data of I-C. Yeh (1998, CC BY 4.0): 1,030 cylinder tests of 428 distinct mixes, 25 exact duplicates removed. A white-box baseline (Abrams' law + ACI 209 strength-gain curve, 7 parameters) is compared with 10 machine-learning models that mirror the MATLAB Regression Learner families. Models are judged on mixes they have never seen (split by mix, not by row, which avoids the data leakage of most published studies) and on age extrapolation. The selected hybrid model - Abrams-ACI trend times a boosted-tree correction - reaches a grouped test RMSE of 4.623 MPa (R2 0.9189) with a split-conformal 90 % prediction interval of +-7.703 MPa, and stays usable when extrapolating in age (6.028 MPa) where pure tree models fail (9.536-8.925 MPa). Differential evolution then searches mixes that meet IS 10262 / IS 456 / IS 16415 rules plus a physics-consistency check; the model proposes 22.6-28.7 % lower material cost and 50.0-58.9 % lower embodied CO2 than a conventional OPC design for M25-M50 - model proposals that need trial batches and cube tests. Honest scope: MATLAB is NOT installed on the build machine - the 14 MATLAB files were written but never run; every number comes from the Python implementation.

Category: Civil Engineering / Concrete Technology / Mix Design (IS 10262, IS 456) / Machine Learning / Optimisation / Sustainable Construction / MATLAB (scripts) with Python

Civil Engineering Project C02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.