One crucial and intricate problem in the education sector that must be dealt with is children who initially enrolled in schools but later dropped out before finishing mandatory primary education. These children are generally referred to as out-of-school children. To contribute to the discuss, this paper presents the development of a robust Multilayer Perceptron (MLP) based Neural Network Model (NN) for optimal prognostic learning of out-of-school children trends in Africa. First, the Bayesian optimization algorithm has been engaged to determine the best MLP hyperparameters and their specific training values. Secondly, MLP-tuned hyperparameters were employed for optimal prognostic learning of different out-of-school children data trends in Africa. Thirdly, to assess the proposed MLP-NN model's prognostic performance, two error metrics were utilized, which are the Correlation coefficient (R) and Normalized root means square error (NRMSE). Among other things, a higher R and lower NRMSE values indicate a better MLP-NN precision performance. The all-inclusive results of the developed MLP-NN model indicate a satisfactory prediction capacity, attaining low NRMSE values between 0.017 - 0.310 during training and 0.034 - 0.233 during testing, respectively. In terms of correlation fits, the out-of-school children's data and the ones obtained with the developed MLP-NN model recorded high correlation precision training/testing performance values of 0.9968/0.9974, 0.9801/0.9373, 0.9977/0.9948 and 0.9957/0.9970, respectively. Thus, the MLP-NN model has made it possible to reliably predict the different patterns and trends rate of out-of-school children in Africa. One of the implications for counselling, among others, is that if every African government is seriously committed to funding education at the foundation level, there would be a reduction in the number of out-of-school children as observed in the out-of-school children data.

Probabilistic parametric functions such as density and distribution functions modeled to depict certain stochastic behaviour are used to express the fundamental theories of reliability engineering. In the existing works of literature, a few probability distribution functions have been well reported. However, selecting and identifying the most suitable distribution functions to reliably model and fit datasets remain. This work examines the application of three different methods for selecting the best function to model and fit measured data. The methods comprise the parametric maximum likelihood estimation, Akaike Information Criteria and the Bayesian Information Criteria. In particular, these methods are implemented on Signal Interference to Noise Ratio (SINR) data acquired over an operational Long Term Evolution (LTE) mobile broadband networks in a typical built-up indoor and outdoor campus environment for three months. Generally, results showed a high level of consistency with the Kolmogorov-Semirnov Criteria. Specifically, the Weibull distribution function showed the most credible performance for radio signal analysis in the three study locations. The explored approach in this paper would find useful applications in modeling, design and management of cellular network resources

In this work, adaptive learning of a monitored real-time stochastic phenomenon over an operational LTE broadband radio network interface is proposed using cascade forward neural network (CFNN) model. The optimal architecture of the model has been implemented computationally in the input and hidden units by means of incremental search process. Particularly, we have applied the proposed adaptive-based cascaded forward neural network model for realistic learning of practical signal data taken from an operational LTE cellular network. The performance of the adaptive learning model is compared with a benchmark feedforward neural network model (FFNN) using a number of measured stochastic SINR datasets obtained over a period of three months at two indoors and outdoors locations of the LTE network. The results showed that proposed CFNN model provided the best adaptive learning performance (0.9310 RMSE; 0.8669 MSE; 0.5210 MAE; 0.9311 R), compared to the benchmark FFNN model (1.0566 RMSE; 1.1164 MSE; 0.5568 MAE; 0.9131 R) in the first studied outdoor location. Similar robust performances were attained for the proposed CFNN model in other locations, thus indicating that it is superior to FFNN model for adaptive learning of real-time stochastic phenomenon.

Propagated electromagnetic signal over the cellular radio communication channels and interfaces are usually highly stochastic and complex with unequal noise variation pattern. This is due to multipath nature of the propagation channels and diverse radio propagation mechanisms that impact the signal strength at the receiver en-route the transmitter, and verse versa. This also makes measurement, predictive modeling and estimation based analysis of such signal very challenging and complex as well. One important and popular parametric modelling and estimation technique in mathematics and engineering science, especially for signal processing applications is the least square regression (LSR). The dominance use and popularity of the LSR approach may be attributed to its simplified supporting theory, relatively fast application procedure and ubiquitous application packages. However, LSR is known to be very sensitive to outliers and unusual stochastic signal data. In this work, we propose the application of weighted least square regression method for enhanced propagation practical field strength estimation modelling over cellular radio communication networks interface. The signal data was collected from a commercial LTE networks service provider. Also, we provide statistical computational analyses to compare the resultant estimation outcome of the weighted least square and the standard least approach. From the result, it is found that the WLSR approach is reliably better the most popular standard least square method. The significance and academic of value of this paper is that our proposed and implemented WLSR method can used as replacement of the standard LSR approach for robust mobile signal processing of future communication system networks.

Mobile phones and handsets enable us to communicate our voice, data and video messages with individuals that are far-off from us. When an active call is initiated by someone using a mobile phone, it is transmitted through a nearby Base Station (BS) transmitter to another BS until the call gets to its intended receiver. Any time a caller initiates and loses a connection to a BS while on conversation, the call is said to be dropped. The initiation and completion of an active call without any form of disconnection or termination is a key service quality parameter in telecommunication system networks. Robust statistical estimation, modelling and characterization of call drop rates is of high importance to the network operators and radio frequency engineers for effective re-planning and performance management process of telecommunication system networks. This work was designed to determine the optimal probability distribution model for drop call rates based on a five week acquired rate of drop calls data sample in the Southern regions of Nigeria.  To accomplish the aim, eight probability distributions namely logistic, log-logistic, normal, log-normal, exponential, Rayleigh, rician and Gumbel max were explored and based on the combined scores of three goodness of fit statistical tests, the log-logistic distribution was found to be the optimal probability distribution for the weekly rate of drop call prognostic analysis. The results could be of immense assistance to radio frequency engineers for optimal statistical modelling and design of cellular systems channels. 

The desire to achieve an adaptive prognostics regression learning processes of physical and empirical phenomenon is a complex task and open problem in radio frequency telecommunication engineering. One key method to solving such complex task or problems is by means of numerical based optimisation algorithms. The Levenberg–Marquardt algorithm (LMA) is an efficient nonlinear parametric machine learning based modelling algorithm with optimal, fast, and accurate convergence speed. This paper proposes and demonstrates the real-time application of the LMA in developing a log-distance like propagation loss model based on received radio strength measurements conducted over deployed long term evolution (LTE) eNodeBs antennas in three different propagation areas. The LTE eNodeB signal propagation areas were selected to reflect typical urban, suburban and rural terrains which represent urban, suburban and rural terrains. The heights of the three eNodeBs are 30, 28 and 32m respectively and each operate at 2.6GHz carrier frequency with 10MHz channel bandwidths. The resultant outcome of the proposed propagation loss modelling using LMA indicates a high approximation efficacy over the popular Gauss-Newton algorithm (GNA) modelling method, which has been used to benchmark the process. Precisely, the developed propagation loss model using LMA method attained lower maximum absolute error (MABE) of 7.73, 14.57and 10.53 for urban, suburban and rural terrains compared to the ones developed by GNA which yielded 15.19, 16.59 and 13.05 MABE values. The improved approximation performance of the LMA over the GNA can be ascribed to its capacity handle multiple free parameters and attain optimum solution irrespective of the selected values of initial guess parameters.

Precise extrapolative mining and analysis of relevant dataset during or after any disease outbreak can assist the government, stake holders and relevant agencies in the health sector to make important decisions with respect to the disease outbreak control and management. While prior works has concentrated on non-stationary long term data, this work focuses on a short term non-stationary and relatively noisy data. Particularly, a distinctive nonparametric machine learning method based kernel-controlled probabilistic Gaussian process regression model has been proposed and employed to model and analyze Covid-19 pandemic data acquired over a period of approximately six weeks. To accomplish the aim, the MATLAB 2018a computational and machine learning environment was engaged to develop and perform the Gaussian process extrapolative analysis. The results displayed high scalability and optimal performance over the commonly used machine learning methods such as the Neural networks, Neural-Fuzzy networks, Random forest, Regression tree, Support Vector machines, K-nearest neighbor and Discriminant linear regression models. These results offer a solid foundation for conducting research on reliable prognostic estimations and analysis of contagious disease emergence intensity and spread.