Fracture of a cell wall causes compressive stress drop in compression of porous metals with unidirectional pores. If such a phenomenon occurs, energy absorption and its efficiency will decrease. This research investigated the dependency of the fractures on initial geometry and the influence of geometric parameters on fractures by statistical analyses. Numerical simulations of compression tests were conducted using the finite element method for eleven specimens with different geometries to efficiently collect data on the occurrence of fractures during compression. A total of eight geometric quantitative parameters were defined for each cell wall which describe the shape of it, its location in the specimen, and the shape of its surrounding cell walls. An additional binary parameter was used to monitor the occurrence of the fracture in the structure. A total of 568 data points were analyzed by a support vector machine and a logistic regression. The prediction results of the support vector machine achieved approximately 0.7 in F1 score, which indicates that the fracture location highly depends on initial geometry. Odds ratios of the logistic regression model show that five out of eight input parameters have a significant influence on the fractures at a significance level of 0.05. Furthermore, it was revealed that cell walls with small relative thickness and relatively small angle, connected to upper and lower cell walls with larger angles, and located near the side surfaces, are more likely to fracture.