The study of phase transitions from experimental data becomes challenging, especially when little prior knowledge of the system is available. Topological data analysis is an emerging framework for characterizing the shape of data and has recently achieved success in detecting structural transitions in material science such as glass-liquid transition. However, data obtained from physical states may not have explicit shapes as structural materials. We propose a general framework called topological persistence machine to construct the shape of data from correlations in states; hence decipher phase transitions via the qualitative changes of the shape. Our framework enables an effective and unified approach in phase transition analysis. We demonstrate the impact in highly precise detection of Berezinskii-Kosterlitz-Thouless phase transitions in the classical XY model, and quantum phase transition in the transverse Ising model and Bose-Hubbard model. Intriguingly, these phase transitions have proven to be notoriously difficult in traditional methods but can be characterized in our framework without requiring prior knowledge about phases. Our approach is thus expected applicable and brings a remarkable perspective for exploring phases of experimental physical systems.