Deep learning (DL) has been widely proposed for botnet attack detection in Internet of Things (IoT) networks. However, the traditional centralized DL (CDL) method cannot be used to detect the previously unknown (zero-day) botnet attack without breaching the data privacy rights of the users. In this article, we propose the federated DL (FDL) method for zero-day botnet attack detection to avoid data privacy leakage in IoT-edge devices. In this method, an optimal deep neural network (DNN) architecture is employed for network traffic classification. A model parameter server remotely coordinates the independent training of the DNN models in multiple IoT-edge devices, while the federated averaging (FedAvg) algorithm is used to aggregate local model updates. A global DNN model is produced after a number of communication rounds between the model parameter server and the IoT-edge devices. The zero-day botnet attack scenarios in IoT-edge devices is simulated with the Bot-IoT and N-BaIoT data sets. Experiment results show that the FDL model: 1) detects zero-day botnet attacks with high classification performance; 2) guarantees data privacy and security; 3) has low communication overhead; 4) requires low-memory space for the storage of training data; and 5) has low network latency. Therefore, the FDL method outperformed CDL, localized DL, and distributed DL methods in this application scenario.

Deep Learning (DL) is an efficient method for botnet attack detection. However, the volume of network traffic data and memory space required is usually large. It is, therefore, almost impossible to implement the DL method in memory-constrained IoT devices. In this paper, we reduce the feature dimensionality of large-scale IoT network traffic data using the encoding phase of Long Short-Term Memory Autoencoder (LAE). In order to classify network traffic samples correctly, we analyse the long-term inter-related changes in the low-dimensional feature set produced by LAE using deep Bidirectional Long Short-Term Memory (BLSTM). Extensive experiments are performed with the BoT-IoT dataset to validate the effectiveness of the proposed hybrid DL method. Results show that LAE significantly reduced the memory space required for large-scale network traffic data storage by 91.89%, and it outperformed state-of-the-art feature dimensionality reduction methods by 18.92-27.03%. Despite the significant reduction in feature size, the deep BLSTM model demonstrates robustness against model under-fitting and over-fitting. It also achieves good generalisation ability in binary and multi-class classification scenarios.

Nowadays, hackers take illegal advantage of distributed resources in a network of computing devices (i.e., botnet) to launch cyberattacks against the Internet of Things (IoT). Recently, diverse Machine Learning (ML) and Deep Learning (DL) methods were proposed to detect botnet attacks in IoT networks. However, highly imbalanced network traffic data in the training set often degrade the classification performance of state-of-the-art ML and DL models, especially in classes with relatively few samples. In this paper, we propose an efficient DL-based botnet attack detection algorithm that can handle highly imbalanced network traffic data. Specifically, Synthetic Minority Oversampling Technique (SMOTE) generates additional minority samples to achieve class balance, while Deep Recurrent Neural Network (DRNN) learns hierarchical feature representations from the balanced network traffic data to perform discriminative classification. We develop DRNN and SMOTE-DRNN models with the Bot-IoT dataset, and the simulation results show that high-class imbalance in the training data adversely affects the precision, recall, F1 score, area under the receiver operating characteristic curve (AUC), geometric mean (GM) and Matthews correlation coefficient (MCC) of the DRNN model. On the other hand, the SMOTE-DRNN model achieved better classification performance with $99.50\%$ precision, $99.75\%$ recall, $99.62\%$ F1 score, $99.87\%$ AUC, $99.74\%$ GM and $99.62\%$ MCC. Additionally, the SMOTE-DRNN model outperformed state-of-the-art ML and DL models.

In this paper, an optimal model is developed for path loss predictions using the Feed-Forward Neural Network (FFNN) algorithm. Drive test measurements were carried out in Canaanland Ota, Nigeria and Ilorin, Nigeria to obtain path loss data at varying distances from 11 different 1,800 MHz base station transmitters. Single-layered FFNNs were trained with normalized terrain profile data (longitude, latitude, elevation, altitude, clutter height) and normalized distances to produce the corresponding path loss values based on the Levenberg–Marquardt algorithm. The number of neurons in the hidden layer was varied (1–50) to determine the Artificial Neural Network (ANN) model with the best prediction accuracy. The performance of the ANN models was evaluated based on different metrics: Mean Absolute error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), standard deviation, and regression coefficient (R). Results of the machine learning processes show that the FNN architecture adopting a tangent activation function and 48 hidden neurons produced the least prediction error, with MAE, MSE, RMSE, standard deviation, and R values of 4.21 dB, 30.99 dB, 5.56 dB, 5.56 dB, and 0.89, respectively. Regarding generalization ability, the predictions of the optimal ANN model yielded MAE, MSE, RMSE, standard deviation, and R values of 4.74 dB, 39.38 dB, 6.27 dB, 6.27 dB, and 0.86, respectively, when tested with new data not previously included in the training process. Compared to the Hata, COST 231, ECC-33, and Egli models, the developed ANN model performed better in terms of prediction accuracy and generalization ability.

A deep knowledge of how radio waves behave in practical wireless channel is required for effective planning and deployment of radio access networks in urban environments. Empirical propagation models are popular for their simplicity, but they are prone to introduce high prediction errors. Different heuristic methods and geospatial approaches have been developed to further reduce path loss prediction error. However, the efficacy of these new techniques in built-up areas should be experimentally verified. In this paper, the efficiencies of empirical, heuristic and geospatial methods for signal fading predictions in the VHF (Very High Frequency) and UHF (Very High Frequency) bands in typical urban environments are evaluated and analysed. Electromagnetic field strength measurements are performed at different test locations within four selected cities in Nigeria. The data collected are used to develop path loss models based on Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS) and Kriging techniques. The prediction results of the developed models are compared with those of selected empirical models and field measured data. Apart from Egli and ECC-33, the Root Mean Squared Error (RMSE) produced by all other models under investigation are considered acceptable. Specifically, ANN and ANFIS models yielded the lowest prediction errors. However, the empirical models have the lowest standard deviation errors across all the bands. The findings of this study will help radio network engineers: to achieve efficient radio coverage estimation; to determine the optimal base station location; to make a proper frequency allocation; to select the most suitable antenna and to perform interference feasibility studies.