Heusler alloys, full and half-, thanks to their high versatility of compositions as well as their very interesting properties, are good candidates for thermoelectric applications. In the Heusler family, quaternary alloys also exist and allow to further increase the chemical diversity and so one to achieve more complex properties. However, due to the high number of combinations, traditional screening methods are not effective to target relevant compounds. To accelerate this research, it is advantageous to use machine learning methods. In our project, we are looking for new promising quaternary Heusler compounds screened within a dataset of 24 selected elements. First, a database of calculated thermodynamic, electronic and magnetic properties, obtain from DFT calculations (Density Functional Theory) on binary and ternary compounds was constructed. Then, a supervised learning with the neural network model was built to predict the enthalpy of formation and the density of state at the Fermi level (metallic or semiconductor character) of quaternary Heusler compounds. Our model presents comparable or superior performance than the state of art and allow to identify promising compounds among the 24^4 possible configurations of our dataset.