The integration of machine learning (ML) approaches in sensor-based applications in the field of pervasive computing is becoming increasingly prominent due to the increasing number of sensor-based applications in general and the continuous adaption of ML approaches to new domains. Several ML models are used within a processing pipeline that operates on the same sensor data. Still, the cloud computing approach is a straightforward solution where all sensor data is sent to the cloud before processing, which is inefficient according to resource utilization. Appropriate management of the different processing tasks for ML models enhances resource utilization. This paper proposes an architecture for resource-aware classification empowered by an ML model management (MLMM) framework and a distributed data stream management system (DDSMS). First, the classification pipeline is decomposed and implemented as data stream operators. Second, ML models are retrieved from an MLMM framework considering preprocessing, segmentation, and feature alignment to enable an effective redundancy elimination. Finally, the classification pipeline is deployed using resource-aware operator placement optimization. The evaluation results on a real-world scenario of a sensor-based activity classification pipeline for dairy cows show that our approach can reduce network utilization up to 98.9%.

Monitoring systems assist farmers in monitoring the health of dairy cows by predicting behavioral patterns (e.g., lying) and their changes with machine learning models. However, the available systems were developed either for indoors or for pasture and fail to predict the behavior in other locations. Therefore, the goal of our study was to train and evaluate a model for the prediction of lying on a pasture and in the barn. On three farms, 7–11 dairy cows each were equipped with the prototype of the monitoring system containing an accelerometer, a magnetometer and a gyroscope. Video observations on the pasture and in the barn provided ground truth data. We used 34.5 h of datasets from pasture for training and 480.5 h from both locations for evaluating. In comparison, random forest, an orientation-independent feature set with 5 s windows without overlap, achieved the highest accuracy. Sensitivity, specificity and accuracy were 95.6%, 80.5% and 87.4%, respectively. Accuracy on the pasture (93.2%) exceeded accuracy in the barn (81.4%). Ruminating while standing was the most confused with lying. Out of individual lying bouts, 95.6 and 93.4% were identified on the pasture and in the barn, respectively. Adding a model for standing up events and lying down events could improve the prediction of lying in the barn.

The recent evolution of the Internet of Things into a cyber-physical reality has spawned various challenges from a data management perspective. In addition, IoT platform designers are faced with another set of questions. How can platforms be extended to smoothly integrate new data management functionalities? Currently, data processing related tasks are typically realized by manually developed code and functions which creates difficulties in maintenance and growth. Hence we need to explore other approaches to integration for IoT platforms. In this paper we cover both these aspects: (1) we explore several emerging data management challenges, and (2) we propose an IoT platform integration model that can combine disparate functionalities under one roof. For the first, we focus on the following challenges: sensor data quality, privacy in data streams, machine learning model management, and resource-aware data management. For the second, we propose an information-integration model for IoT platforms. The model revolves around the concept of a Data-Sharing Market where data management functionalities can share and exchange information about their data with other functionalities. In addition, data-sharing markets themselves can be combined into networks of markets where information flows from one market to another, which creates a web of information exchange about data resources. To motivate this work we present a use-case application in smart cities.