This thesis will study how machine learning can be used to predict the temperature range in which the austenite phase remains stable in steel materials but more specifically steel alloys. Li et al. (2024) states that the state of austenite grains has an essential role in the microstructure transformation and mechanical properties of alloy steel during heat treatment.  Austenite is a phase in metal processing that serves as the foundation for various heat treatments and is characterized by its face-centered cubic crystal structure (FCC). Bongao et al. (2026) investigate how controlling the stability of austenite during heat treatment is important to achieve the desired microstructure and material properties. Predicting the temperature stability of  austenite is important for optimizing heat treatment parameters and controlling the result microstructure and mechanical properties of the alloys. 

This study utilizes a thermodynamic equilibrium dataset containing 3,840 unique steel alloy compositions which each is characterized by eight elements. These elements consist of aluminum (Al), carbon (C), iron (Fe), manganese (Mn), molybdenum (Mo), niobium (Nb), silicon (Si), and vanadium (V). The dataset consists of approximately 3 million rows of phase equilibrium data spanning temperatures between 773°C  and 1073°C . 

There are two targeted variables that were engineered and analyzed, the first was the austenite dominant range. This temperature range where austenite has the highest volume fraction among all phases. The second was the austenite threshold range and this is when the threshold range has a volume fraction exceeding 90\%. The machine learning algorithms that are implemented and compared are Linear Regression and Random Forest Regression. 

To find the austenite dominant range prediction, Random Forest achieved a R² score of 0.9989 with a RMSE of 2.28°C. The Random Forest outperformed Linear Regression which had a R² = 0.8792 and RMSE = 23.81°C. The results were observed for the threshold-based analysis where Random Forest demonstrated R² = 0.9984 and RMSE = 1.75°C. The analysis revealed that manganese (54.5\%) and aluminum (24.6\%) are the influential factors in determining austenite stability for the dominant range model. Carbon has a positive coefficient in Linear Regression which is consistent with its known role as an austenite stabilizer. 

The results from the study demonstrate that machine learning can predict austenite temperature stability based on alloys composition hence providing a tool for materials engineers to optimize steel alloy design without extensive experimental testing. A prompt-line application for prediction was developed to enable practical use of the trained models.