School of Computing and IT (CP)http://repository.mut.ac.ke:8080/xmlui/handle/123456789/29302024-03-28T14:04:57Z2024-03-28T14:04:57ZHarnessing our ICT Skill Set and Research Efforts for Sustainable DevelopmentMuketha, Geoffrey M.http://repository.mut.ac.ke:8080/xmlui/handle/123456789/61462024-02-17T11:23:36Z2022-06-01T00:00:00ZHarnessing our ICT Skill Set and Research Efforts for Sustainable Development
Muketha, Geoffrey M.
For some time now, ICTs have become ubiquitous, making them a common place phenomenon in modern society. Almost all electronics are ICT enabled to date, with most people who can read and write having interacted with ICTs one way or the other. For example, mobile phones which have changed the way we think, socialize and do business, are ICT-enabled. According to Statista, a German Company specializing in market and consumer data, The number of mobile subscriptions in Kenya rose from 0.13 million in the year 2000 to 61.41 million in 2020. This makes sense when you consider that many Kenyans walk with two or more mobile devices wherever they go.
Countries that have had the highest infusion of ICT in their sectors have reaped the most benefits. These benefits can generally be seen alongside the developed vs developing countries divide. ICTs are seen as a powerful solution to the problems affecting developing countries. Indeed, ICTs have the potential of making developing countries to catch up and even overtake developed countries much faster than originally thought.
This paper challenges us to harness ICT skill set through: 1. enhancement of our computing curricula by thinking through how they are designed and implemented. 2. Enhancement of research efforts by addressing four key problems that characterize research in this sector so as to come up with novel and sustainable solutions to challenges that affect our developing economies.
2022-06-01T00:00:00ZEmpirical Evaluation of Adaptive Optimization on the Generalization Performance of Convolutional Neural NetworksWanjau, Stephen K.Wambugu, Geoffrey MOirere, Aaron M.http://repository.mut.ac.ke:8080/xmlui/handle/123456789/54972024-02-17T11:19:55Z2021-10-01T00:00:00ZEmpirical Evaluation of Adaptive Optimization on the Generalization Performance of Convolutional Neural Networks
Wanjau, Stephen K.; Wambugu, Geoffrey M; Oirere, Aaron M.
Recently, deep learning based techniques have garnered significant interest and popularity in a variety of fields of research due to their effectiveness in search for an optimal solution given a finite amount of data. However, the optimization of these networks has become more challenging as neural networks become deeper and datasets growing larger. The choice of the algorithm to optimize a neural network is one of the most important steps in model design and training in order to obtain a model that will generalize well on new, previously unseen data. In deep learning, adaptive gradient optimization methods are mostly preferred for supervised and unsupervised tasks. First, they accelerate the training of neural networks and since mini batches are selected randomly and are independent, an unbiased estimate of the expected gradient can be computed. This paper examined six state-of-the-art adaptive gradient optimization algorithms, namely, AdaMax, AdaGrad, AdaDelta, RMSProp, Nadam, and Adam on the generalization performance of convolutional neural networks (CNN) architecture that are extensively used in computer vision tasks. Experiments were conducted giving comparative analysis on the behaviour of these algorithms during model training on three large image datasets, namely, Fashion-MNIST, Kaggle Flowers Recognition and Scene classification. The results show that Adam, Adadelta and Nadam finds the global minimum faster in the experiments, have a better convergence curve, and higher test set accuracy in experiments using the three datasets. These optimization approaches adaptively tune the learning rate based only on the recent gradients; thus, controlling the reliance of the update on the past few gradients.
2021-10-01T00:00:00ZA Novel Hybrid Deep Learning Model for Early Detection of Diabetic RetinopathyWanjau, Stephen K.Muketha, Geoffrey M.http://repository.mut.ac.ke:8080/xmlui/handle/123456789/51342021-09-28T14:57:02Z2021-01-01T00:00:00ZA Novel Hybrid Deep Learning Model for Early Detection of Diabetic Retinopathy
Wanjau, Stephen K.; Muketha, Geoffrey M.
Diabetic retinopathy is one of the most frightening complications of diabetes mellitus affecting the working-age population worldwide leading to irreversible blindness if left untreated. A major challenge is early detection, which is very important for treatment success. Presently, detecting diabetic retinopathy is a time-, effort-, and cost-consuming manual process where ophthalmologists identify diabetic retinopathy by the presence of lesions associated with the vascular abnormalities caused by the disease. However, the expertise and equipment needed are often lacking in places where the rate of diabetes in the populace is high and detection is most wanted. As the number of persons diagnosed with diabetes continues to increase, the infrastructure required to prevent diabetes-induced blindness will become even more important. The need for an automated approach to diabetic retinopathy screening has long been recognized, and several recent efforts have made good progress using image classification, pattern recognition, and machine learning. With color fundus photography as input, we proposes a novel approach to detection of diabetic retinopathy based on deep learning techniques ideally resulting in a model with realistic clinical potential. A hybrid configuration of a 2-dimensional Convolutional Neural Network (CNN) and Extreme Learning Machine (ELM) is proposed to leverage feature representation of CNN and fast classification learning of ELM. A publicly available benchmark dataset consisting of 35,126 retina scan images that are resized into 224x224 pixels are used to train, test and evaluate the proposed model. To measure the performance of the proposed model, the accuracy, precision, sensitivity, F1-score, and Cohen’s kappa score were determined. Model results reveal that the CNN-ELM approach achieved an accuracy of 94.72%, 92.71% precision, 0.7834 sensitivity, an F1-score of 0.8492 and an average Kappa-score of 0.6792 for the multi-class classification. These results demonstrate the proposed hybrid model’s ability to detect and classify diabetes retinopathy.
2021-01-01T00:00:00ZImplementation of a Structural Complexity Metrics Tool for Sassy Cascading Style Sheets (SCMT-SCSS)Ndia, John G.Muketha, Geoffrey M.Omieno, Kelvin Kabetihttp://repository.mut.ac.ke:8080/xmlui/handle/123456789/44072024-02-17T11:17:30Z2020-01-01T00:00:00ZImplementation of a Structural Complexity Metrics Tool for Sassy Cascading Style Sheets (SCMT-SCSS)
Ndia, John G.; Muketha, Geoffrey M.; Omieno, Kelvin Kabeti
The development of metrics tool is a basic requirement for the defined software metrics to be acceptable in the software industry. There are several metrics proposed over the years without tool support and this trend cannot be tolerated. Therefore, a metrics tool to automate the collection and analyse the defined Sassy CSS (SCSS) metrics has been developed and is herein referred as Structural Complexity Metrics Tool for Sassy Cascading Style Sheets (SCMT-SCSS). The tool was validated by using 4 SCSS files, and a convenient sample of 21 students who were trained in SCSS language. The results indicate that the tool computes metrics in a much less time than manual computation. The subjects rated the tool as very suitable for the tasks assigned, very accurate in computations of metrics values and very operable.
2020-01-01T00:00:00Z