Nowadays, special focus on intelligent and context-aware data management is given in the intersection between Internet of Things (IoT) and Edge Computing (EC), where humongous volumes of data are collected and transferred towards loud data centers. Edge nodes act as intermediaries between IoT devices and data centers, thus, their behaviour and performance are critical for supporting high-quality, data-driven pervasive services to end-users. In this paper, we propose a context-aware strategy for data selectivity at the network edge that deals with the challenges of limited computational and storage resources. Our goal is to efficiently learn the access patterns of incoming data-driven tasks and predict which data are relevant, thus, should be stored in nodes’ local datasets. Tasks are associated with processing activities over local datasets directly indicating the data sub-spaces required to be accessed. We adopt one-class classification statistical learning to identify the data sub-spaces requested by tasks while eliminating outlier requests. In turn, we develop an uncertainty-driven mechanism based on fuzzy inference rules that determines a data filter mechanism capable of predicting the future relevant data sub-spaces. We analytically describe our mechanism and comprehensively evaluate and compare it against baselines and models found in the literature showcasing its applicability in pervasive edge computing environments.