Big Data and IoT for real-time miscarriage prediction A clustering comparative study

Abstract

Abstract Sensors and mobile phones are becoming very useful to gather a huge amount of data rapidly and to understand the behaviour of the human. Today, the real challenge is how to benefit from the use of Big Data performant tools and machine learning algorithms to get the desired value. The present paper presents a comparative study of clustering algorithms (Kmeans, Bisecting Kmeans and Gaussian mixture) for a real time miscarriage prediction. Wearable healthcare sensors ((heart rate sensor, temperature sensor and activity sensor) and mobile phone are used for gathering real time data about the pregnant women. Sensors are managed using IoT technologies such as Raspberry Pi to collect and process data in real time. Prediction’s results are sent to the doctor through a mobile phone created and the pregnant woman receives recommendations based on her behaviour. Our study compares the performance and the efficiency of the predictive models created by the three algorithms, including time to build the model, clusters distribution and centres definition. We evaluate models using the Internal clustering validation silhouette method. The dataset generated and analyzed during the current study is available on GitHub platform via the following link: https://github.com/hibaasri/Miscarriage-Prediction . Databricks platform and Spark are used to analyze data and build models.