Ensemble learning prediction of transmittance at different wavenumbers in natural hydroxyapatite

Material engineering-based research has often relied so much on tedious human experiments for generating specific engineering properties with a major draw-back of high time demand that can span between an hour and days. Hence to deviate from the usual paradigm, we provide an alternative approach which employs artificial intelligence (AI) based ensemble learning methods for predicting the degree of transmittance for a range of wavenumbers of infrared radiation through hydroxyapatite (HAp) samples. The effective samples (transmittance and wavenumber) were passed as input to the predictive systems. For this, we trained two ensemble learning methods: Extreme Gradient Boosting (XGBoost) and Random Forest on variants of HAp (density and time variations), while considering a fixed amount of 10,000 base estimators. The results show that Random Forest marginally outperforms the XGBoost in the testing phase but requires a much longer computing time. However, XGBoost is much faster than the Random Forest. Furthermore, the examined ensemble learning models yielded a coefficient of determination (R² > 0.997): which are in close agreement with experimental data, depicting an excellent generalization capacity. Additionally, the examined ensemble learning models showed a significant ≥ 99.83% decrease in computational complexity relative to the time spent when generating the experimental data. Overall, the use of ensemble learning models is very important for validating material engineering properties.