Duplex stainless steels (DSSs) have an attractive combination of excellent mechanical properties as well as corrosion resistance, making them to be a suitable choice in applications within severe environments, e.g. chemical engineering plant, nuclear power plant, etc. However, DSSs are reported to be quite sensitive to be brittle during the ‘475 oC embrittlement’ which limits its service temperature. This problem exists over sixty years in the DSSs community while the solution to solve this problem is still not possible to provide. In a recent work, it is found that ‘475 oC embrittlement’ can be directly related to the hardness of ferrite. This work established a machine learning (ML) model to predict the micro-hardness evolution of ferrite with time dependence at various temperatures. The data are collected from the open literatures as well as the experimental data of the commercial DSSs (2101, 2205, 2304, 2507, 3207) which were heat treated at different conditions. Five ML models, i.e. linear regression model (LR), regression tree (RT), support vector machine (SVM), Gaussian process regression (GPR), and ensemble tree (ET) were used to train the model. The established database was randomly separated 80% for training and 20% for testing. Furthermore, the unseen data was used to validate the current ML hardness model. Utilizing the established ML model, the effects of the alloying elements, e.g. Cr, Ni, Mn, and N, etc. as well as the heat treatment conditions, i.e. aging time and temperature are further investigated. Finally, lab-scale DSS alloys were prepared to investigate the effect of Ni content on the ferrite hardness evolution, and to benchmark the ML modelling capability. The obtained methodology of ML modelling aims to provide a new methodology to establish ML database using a combination of literature data as well as the experimental measurement, and to support the understanding of the mechanism of low temperature embrittlement of DSSs.