The research of flow fields is often limited by the high computational cost of generating, storing, and processing the database, which can either come from computational fluid dynamics (CFD) simulations or experiments. A refined mesh is needed to obtain accurate CFD solutions, to which a high computational cost is associated. As for experimental results, they are very costly and time-consuming. In this context, reduced order models (ROMs) came up as a solution to solve industrial and academic applications. ROMs reduce the data dimensionality and can be used to speed up numerical simulations or to predict the evolution of experimental measurements. Fluid-mechanics problems usually apply ROMs in order to obtain spatio-temporal physical patterns that describe the fluid motion. The large number of variables involved in reactive flows encourages the need of extracting the main features representing the system and the use of ROMs to reduce data-dimensionality. On the one hand, Proper Orthogonal Decomposition (POD) [1] is a well-known technique that extracts physical patterns in fluid-mechanics problems while reducing the data dimensionality. It extracts orthogonal modes sorted by their contribution to the total energy of the system to accurately model the ROM. On the other hand, the ROMs resulting from Dynamic Mode Decomposition [2], have recursively been used to identify flow instabilities. Furthermore, Le Clainche and Vega presented an extension of this method, the Higher Order DMD (HODMD) [3], for the analysis of complex flows (turbulence, multi-scale, etc) [5,6]. In this work, we aim to compare the conventional techniques of POD and HODMD to obtain ROMs with the novel techniques of Autoencoders (AEs) and Vector Quantization Principal Component Analysis (VQPCA) [4], in terms of compression of the original dataset and ability to extract physical features on a complex flow defined by the synchronized movement of two synthetic jets.