Analyzing Temperature-Dependent Thermal Properties of Biomaterials Using Machine Learning Methods

Abstract

This research utilizes machine learning models ANN, Random Forest, and Decision Tree to predict material properties of Ti-based biomaterials, including Young’s modulus, density, thermal conductivity, and specific heat at various temperatures. Data was sourced using web scraping and Plot Digitizer, validated against literature, and analyzed in Excel. The ANN model achieved strong performance, with R² = 0.980874 for TiAl and R² = 0.997607 for TiCu, effectively predicting density and Young’s modulus but showing deviations in band gap. For TiO2, the ANN model demonstrated solid predictions but struggled with band gap and specific heat accuracy. Random Forest yielded high accuracy for TiAl (R² = 0.998168) and TiO2 (R² = 0.9994) and its ability to generalize well and capture complex relationships in the data makes it the most reliable method for this study. The Decision Tree model accurately predicted specific heat and Young’s modulus for TiAl (R² = 0.993841) and captured trends in TiCu but showed deviations in band gap and thermal conductivity. These results underline the predictive potential of these models while highlighting areas for refinement.