Algorithm 1 Proposed localization procedure.

<Offline stage> Input     Training set ${\mathbf{R}}^{\prime }$: The set of dual-band RSS vector, ${\mathbf{r}}_i=[r_i^{2.4},r_i^{5.2}]$,     and ground truth distance $d_i$ for each sample     Validation set ${\mathbf{V}}^{\prime }$: The set of validation data, which are same kind of data as,     but not included in the training set Initialize a neural network described in Figure 4 with random weights, biases, and normalization parameters. for all training data grouped by mini-batch size of $K_B$, do     Forward calculation for the NN     The output $y_i$ is a result of the last layer of the NN     Calculate the average MSE between $y_i$ and $d_i$ in the mini-batch     Update parameters of the NN, by using back–propagation end for Find best–fit parameters of the NN that minimize MSE for the validation set Output Trained NN as a deep regression model   <Online stage> Input     Test data location of N APs and N RSS vector Set a confidence threshold $T{h}_c$ for each i-th AP do     Predict ${\widehat{d}}_i$ using the trained NN end for Find target position $\widehat{\mathbf{x}}$, using Equations (5)–(7) Output: Position estimate of target device $\widehat{\mathbf{x}}={[\widehat{x},\widehat{y}]}^{\mathrm{T}}$