• Neural networks outperform traditional regression methods in predicting material properties.
• Feature selection and data normalization are critical for model accuracy.
• The proposed model can reduce experimental testing time by up to 70%.
• The methodology is applicable to a wide range of material systems.