Résumé : Indoor localization is the process of determining the location of a person or object inside a building. Potential usage of indoor localization includes navigation, personalization, safety and security, and asset tracking. Commonly used technologies for indoor localization include WiFi, Bluetooth, RFID, and Ultra-wideband. Among these, WiFi's Received Signal Strength Indicator (RSSI)-based localization is preferred because of widely available WiFi Access Points (APs). We have two main contributions. First, we develop our method, 'IndoorGNN' which involves using a Graph Neural Network (GNN) based algorithm in a supervised manner to classify a specific location into a particular region based on the RSSI values collected at that location. Most of the ML algorithms that perform this classification require a large number of labeled data points (RSSI vectors with location information). Collecting such data points is a labor-intensive and time-consuming task. To overcome this challenge, as our second contribution, we demonstrate the performance of IndoorGNN on the restricted dataset. It shows a comparable prediction accuracy to that of the complete dataset. We performed experiments on the UJIIndoorLoc and MNAV datasets, which are real-world standard indoor localization datasets. Our experiments show that IndoorGNN gives better location prediction accuracies when compared with state-of-the-art existing conventional as well as GNN-based methods for this same task. It continues to outperform these algorithms even with restricted datasets. It is noteworthy that its performance does not decrease a lot with a decrease in the number of available data points. Our method can be utilized for navigation and wayfinding in complex indoor environments, asset tracking and building management, enhancing mobile applications with location-based services, and improving safety and security during emergencies.