With the development of Internet of Things (IoT) and machine learning technologies, mobile geographic information systems (GISs) have developed rapidly. Moreover, mobile GIS applications serve all walks of life including remote sensing, geological disaster management, and decision support systems. This article discusses the main development methods of the Android system for mobile GIS, analyzes the characteristics of different development methods, and mainly introduces the technology of developing mobile GIS based on free and open-source software (FOSS) framework. Finally, we present a data collection framework for an Android application development, based on QGIS, QFiled, GeoServer, PostgreSQL, and GeoPackage. The mobile GIS can collect important data. Furthermore, the data collection framework uses a data aggregation technique to filter and remove redundant data. Machine learning approaches are integrated in the GIS to make it intelligent. The application, in the Xishan mining area of Taiyuan, proves that the proposed framework can complete the collection and storage of geological disaster data, which has certain practical significance. Our experimental results show that the data aggregation method is approximately 42.3–44.09 percent (training times) more efficient than the no aggregation approach. Moreover, the attention network may produce an additional overhead in the prediction process, depending on the model. This overhead is observed between 0.58% and 2.83% for the LSTM model.