Wireless connections are more and more used in different applications and in public areas for services to consumers but also for handling (sometimes) sensitive communications (for instance in railway systems or for remote video monitoring systems). Such systems can have to face different kind of attacks that target the behind service. Our work aims to detect, as soon as possible and online, attacks that can occur on wireless networks, to be able to react very quickly. In this paper, we present some results of data analysis methods, on Wi-Fi signals, to differentiate the ones with attacks from the ones without. This study focuses on low power jamming attacks with a slight or even no impact on Wi-Fi communications. This is more challenging than detecting high power jamming attacks which have already been addressed in the literature. Being able to detect a low impact attack is a crucial issue in a global security strategy, making it possible to launch countermeasures before the interruption of the communication. The Wi-Fi bands are also in the ISM frequencies, making the environment complicated to analyze. Clustering methods such as Agglomerative Hierarchical Clustering are used to identify some clusters and then to map them to the real classes (with or without attacks). A deep analysis of the clusters obtained in a dataset acquired in uncontrolled conditions is carried out. This is done in order to understand what is responsible of the clustering assignment of the different points and to extract the clusters which can be used to design a detection attack strategy.