DLULIB: a multi-floor WiFi Fingerprinting dataset for indoor positioning in Dalat University library

In this paper a multi-floor WiFi Fingerprinting dataset collected in the library of Dalat University, Vietnam, is presented to support indoor positioning research in complex real-world environments. The dataset was constructed from three floors of the building, which cover open reading areas and corridors under various conditions. Using multiple smartphones operated by different participants and participant-facing directions, a total of 19,500 WiFi fingerprints were collected at 650 reference points (RPs) on three floors. Each fingerprint includes Received Signal Strength (RSS) values collected from 17 access points. It also contains the spatial two-dimensional coordinates of the corresponding RPs and metadata such as device type, participant identifier, and participant facing direction. The dataset includes training (offline phase) and testing (online phase) subsets, which are provided in structured Excel files for each floor. This organization enables flexible use of the dataset for signal analysis, preprocessing evaluation, and algorithm development. Simple-to-complex methods, including k-nearest neighbors, convolutional neural networks, and transformers, are implemented to illustrate dataset usability by considering baseline mean positioning errors. The dataset is publicly accessible through a data repository and can be used for benchmarking indoor positioning approaches, analysis of device heterogeneity, and analysis of signal fluctuation in multi-floor environments.