Design of liquid filled tank remains an important challenge in various fields such as transport or energy production. Indeed, sloshing in such container could generate critical situations such as loss of stability of a vehicle, fatigue or even structural failure. In order to mitigate these issues, a classical approach consists in immersing baffles in the tank. The efficient design of these devices remains a difficult task since the sloshing phenomenon is non-linear and the numerical modeling of the fluid-structure interaction computationally expensive. This is particularly the case when the associated numerical solver is integrated in a global optimization loop which requires a large number of function evaluations. This first work focuses here on a fully linear approach for the sloshing modeling and for the structure. It aims at proposing an efficient methodology for optimal pre-design of baffles in a tank subjected to sloshing. Two main ingredients are considered to reach this goal: (1) a robust and efficient numerical solver based on eXtended Finite Element Method (XFEM) to simulate the fluid-structure interaction and (2) a Bayesian Optimization algorithm based on Gaussian Process to optimize baffle's shape and position. The numerical solver provides accurate solutions (pressure, velocity, free surface elevation and structure displacements) using XFEM to model the discontinuity of the pressure field at the fluid-structure interface on the frequency domain. The proposed approach allows to reduce the number of built operators for each new set of design parameters and so it leads to a significant reduction of the computational cost to get the solutions along the parameters changes. An additional reduction of the computational cost is obtained by using a Constrained Bayesian Optimization algorithm which uses iteratively enriched surrogate models to approximate the objective function and the constraints. This whole strategy allows to efficiently find optimal baffle's design minimizing the sloshing amplitude on the free surface with respect to a maximum force acting by the fluid on the structure. It is applied on 3D academic test cases such as parallelepiped tanks with one immersed baffle. After presenting and validating the numerical solver in terms of accuracy and computational cost, the efficiency of the optimization strategy is illustrated on examples including up to 6 design parameters.