Abstract: Detonation cell width λ is an important parameter in quantifying explosion risk of a flammable gas mixture. Generally, the ratio between λ and characteristic chemical reaction zone width δ is considered as a function of a dimensionless activation energy and a dimensionless temperature. As a step further than the regression for experimental data with these two dimensionless parameters as independent variables and with the logarithm of λ/δ as the dependent variable, a dimensionless pressure was introduced as the 3rd independent variable in regression. Meanwhile, in consideration of the disadvantages of classical parametric regression methods, support vector regression method based on machine learning was applied to fit the data. The comparison among the regression results shows that, compared with the 2-variable model and parametric regression method, the 3-varaible model with support vector regression method can offer better fitness to experimental data as well as higher accuracy in predicting detonation cell width for flammable gas mixtures with different initial conditions.