Specification and forecast ionospheric parameters, such as ionospheric
electron density (Ne), have been an important topic in space weather and
ionosphere research. Neural networks (NNs) emerge as a powerful modeling
tool for Ne prediction. However, heavy manual attention costs time to
determine the optimal NN structures. In this work, we propose to use
neural architecture search (NAS), an automatic machine learning method,
to address this problem of NN models. NAS aims to find the optimal
network structure through the alternated optimization of the
hyperparameters and the corresponding network parameters. A total of
16-year data from Millstone Hill incoherent scatter radar (ISR) are used
for NN models. One single-layer NN (SLNN) model and one deep NN (DNN)
model are trained with NAS, namely SLNN-NAS and DNN-NAS, for Ne
prediction and compared with their counterparts without NAS from
previous studies, denoted as SLNN and DNN. Our results show that
SLNN-NAS and DNN-NAS outperformed SLNN and DNN, respectively. NN models
can reveal more finer details than the empirical ionospheric model
developed using traditional data fitting approaches. DNN-NAS yields the
best prediction accuracy measured by quantitative metrics and rankings
of daily pattern prediction. The limited improvement of NAS is likely
due to the network complexity and the limitation of fully connected NN
without a memory mechanism.