Smart and efficient application of DL algorithms in IoT devices can improve operational efficiency in healthcare, including tracking, monitoring, controlling, and optimization. In this paper, an artificial neural network (ANN), a structure of deep learning model, is proposed to efficiently work with small datasets. The contribution of this paper is two-fold. First, we proposed a novel approach to build ANN architecture. Our proposed ANN structure comprises on subnets (the group of neurons) instead of layers, controlled by a central mechanism. Second, we outline a prediction algorithm for classification and regression. To evaluate our model experimentally, we consider an IoT device used in healthcare i.e., an insulin pump as a proof-of-concept. A comprehensive evaluation of experiments of proposed solution and other classical deep learning models are shown on three small scale publicly available benchmark datasets. Our proposed model leverages the accuracy of textual data, and our research results validate and confirm the effectiveness of our ANN model.