PHD-School of Engineering And Technology

Permanent URI for this community

https://repository.ku.ac.ke/handle/123456789/248

This collections contains bibliographic information and abstracts of PHD theses and dissertation in the School of Engineering And Technology held in Kenyatta University Library

Browse

Now showing 1 - 11 of 11

Factors that determine the performance of technology-Based firms in kenya
(2011-08-02) Kinoti, Kaburu Franklin
This study aimed to investigate, in a knowledge based framework, the determinants of performance of Kenyan technology-based firms focusing on the role of human capital, social capital and traditional firm-level characteristics on firm new knowledge acquisition through R&D and technology acquisition (innovation inputs) and the transformational process leading to innovation output and firm performance. The empirical analysis focused on a sub-sample of 320 high and medium-high technology firms drawn randomly from a population of 772 firms located in Nairobi. The sample population was stratified using seven technology-based industrial sectors and three employment size bands. Data collection was done using a self-administered structured questionnaire and analysis done using SPSS version 11.5. The innovative capability was analyzed by use of descriptive statistics while the relationship linking investment in new knowledge, innovation and productivity was established by separately modeling the & terminants of technology acquisition, R&D, innovation output and firm productivity. The study employed logit analysis for R&D decision and innovation outcome, probit analysis for technology acquisition, Tobit for R&D intensity and augmented Cobb Douglass production function for productivity. Empirical evidence revealed that the firm's innovations were largely incremental and that the innovative capability of the firms was largely inadequate. The results demonstrated that size and exporting variables were significant predictors of R&D decision, technology acquisition, innovation output and firm economic performance but not R&D intensity. Of the two innovation inputs, only technology acquisition increased probability of introducing innovations. On the other, hand innovation output contributed significantly to increased firm performance as measured in terms of value-added. Human capital variables had significant positive effect on all the dimensions under study while the role of social capital was multifaceted in its effect. While general linkages with competitors and other institutions had significant influence on the firms to invest in new knowledge, only linkages with customers had significant and positive effects on the likelihood of the firms to innovate. On the productivity side linkages with competitors joined linkages with customers in increasing valued added. Lastly the following conclusions can be drawn. First, it appears that R&D directly contributed to higher firm performance during the study period by increasing the absorptive capacity and not indirectly through innovation propagation as the main hypothesis posited, at least for Kenyan technology-based firms. Thus policies geared towards increasing the capability to transform R&D activities into commercial innovations would significantly increase the innovative performance of the firms. Second, since exporting large firms that had higher level of scientific and technical workforce and qualified managers, had cooperated with customers and competitors, had invested in R&D activities and technology acquisition and had launched new or improved products processes to the markets performed better than those that did not, public policies meant to stimulate increased firm growth and export promotion deepening access to qualified human resources promotion of linkages between firms and other institutions; promotion of in-house R&D and external acquisition of technology in both embodied and disembodied form should have positive results in terms of the overall performance of the firms.
Forecasting Seasonal Stream Flow in Athi River Basin Using Global and Regional Climate Predictors
(Kenyatta University, 2011-11) Obiero, Cllfford Clement Nyaberi
Extreme weather events are associated with floods and drought which affect infrastructure, food security, water availability, sanitation and hydro power generation. The Athi river basin is prone to such extreme weather events. This study examines the spatial and temporal hydrological characteristics of the Athi river basin, their teleconnections with EI Nino/Southern Oscillation (ENSO) and the Indian Ocean Dipole (lOD) and assesses the potential use of derived teleconnections for stream flow forecasting. The Principal Component analysis (PCA) method was used to design the minimum rainfall and stream flow networks for the basin. Spatial and temporal PCA modes were used to study the spatial characteristics of seasonal rainfall and stream flow. Trends were studied using graphical and Spearman rank correlation methods while periodicity was investigated using wavelet analysis. PCA and composite analysis were used to map linkages among extreme seasonal rainfall/stream flow patterns and ENSO and [00 evolution phases for the 'long' and 'short' rainy seasons. Simple linear correlation and Canonical Correlation Analysis (CCA) were used to delineate interlinkages among rainfall, stream flow, ENSO and lOD. The potential of deriving seasonal rainfall and stream flow forecasts was investigated using the step-wise regression and the non-parametric seasonal forecasting (NSFM) models. PCA delineated the Athi river basin into six rainfall and three streamflow homogenous zones. The spatial characteristics of seasonal rainfall and streamflow showed a strong influence of the Inter Tropical Convergence Zone (lTCZ), land sea mesoscale circulation system, orography and land use systems. The PCA based areal rainfall and streamflow indices and communality analysis were used to determine the best representative stations for the homogenous zones. PCA (T-mode) delineated most of the wet /high flow and dry/low flow years observed from historical records including 1997/1984 which are some of the wettest/driest years on record in the basin. Although, the results showed significant rainfall and stream flow trends at some locations, it was difficuIt to associate the observed trends to climate change, due to limited data length. Wavelet analyses showed significant peaks centered at 2-3, 5-7 and 10-12 years which may be associated with Quasi-Biennial Oscillation, ENSO, solar cycles and decadal variability modes. Linkages between various modes of ENSO were also delineated including "El Nino Modoki" which is a manifestation of EI Nino associated with dry instead of the normal wet conditions in the basin. Strong linkages of seasonal rainfall and stream flow with ENSO and 10D phases with time lags of 7-9 months and lag correlations of 0.8 - 0.9 were obtained. The composite results indicated that the wet/high flow and dry/low flow conditions in the basin could be associated with the evolution of ENSO and IOD phases. CCA method delineated the major Indian Ocean Sea Surface Temperature (SST) modes, which are associated with extreme March-May and September-November seasonal rainfall and streamflow. Regression and NSFM models showed good predictions skills for low and high flows using ENSO and IOD predictors. The results of this study provide tools for prediction, early warning of the extremes episodes, management, planning and operation of water resources systems.
Forecasting seasonal stream flow in Athi River basin using global climate predictors
(2012-04-20) Obiero, Clifford Clement
Extreme weather events are associated with floods and drought which affect infrastructure, food security, water availability, sanitation and hydro power generation. The Athi river basin is prone to such extreme weather events. This study examines the spatial and temporal hydrological characteristics of the Athi river basin, their teleconnections with EI Nifio/Southern Oscillation (ENSO) and the Indian Ocean Dipole (lOD) and assesses the potential use of derived teleconnections for stream flow forecasting. The Principal Component analysis (PCA) method was used to design the minimum rainfall and stream flow networks for the basin. Spatial and temporal PCA modes were used to study the spatial characteristics of seasonal rainfall and stream flow. Trends were studied using graphical and Spearman rank correlation methods while periodicity was investigated using wavelet analysis. PCA and composite analysis were used to map linkages among extreme seasonal rainfall/stream flow patterns and ENSO and IOD evolution phases for the 'long' and 'short' rainy seasons. Simple linear correlation and Canonical Correlation Analysis (CCA) were used to delineate interlinkages among rainfall, stream flow, ENSO and IOD. The potential of deriving seasonal rainfall and stream flow forecasts was investigated using the step-wise regression and the non-parametric seasonal forecasting (NSFM) models. PCA delineated the Athi river basin into six rainfall and three streamflow homogenous zones. The spatial characteristics of seasonal rainfall and streamflow showed a strong influence of the Inter Tropical Convergence Zone OTCZ), land sea mesoscale circulation system, orography and land use systems. The PCA based areal rainfall and streamflow indices and communality analysis were.used to determine the best representative stations for the homogenous zones. PCA (T-mode) delineated most of the wet /high flow and dry/low flow years observed from historical records including 1997/1984 which are some of the wettest/driest years on record in the basin. Although, the results showed significant rainfall and stream flow trends at some locations, it was difficult to associate the observed trends to climate change, due to limited data length. Wavelet analyses showed significant peaks centered at 2-3, 5-7 and 10-12 years which may be associated with Quasi-Biennial Oscillation, ENSO, solar cycles and decadal variability modes. Linkages between various modes of ENSO were also delineated including "El Nino Modokt which is a manifestation of El Nifio associated with dry instead of the normal wet conditions in the basin. Strong linkages of seasonal rainfall and stream flow with ENSO and IOD phases with time lags of 7-9 months and lag correlations of 0.8 - 0.9 were obtained. The composite results indicated that the wetlhigh flow and dry/low flow conditions in the basin could be associated with the evolution of ENSO and IOD phases. CCA method delineated the major Indian Ocean Sea Surface Temperature (SST) modes, which are associated with extreme March-May and September-November seasonal rainfall and streamflow. Regression and NSFM models showed good predictions skills for low and high flows using ENSO and IOD predictors. The results of this study provide tools for prediction, early warning of the extremes episodes, management, planning and operation of water resources systems.
An assessment of factors affecting adoption of appropriate technologies by rural women in Meru and Kiambu districts, Kenya
(2012-06-08) Mburugu, Keren G.
Appropriate technology (AT) has been recognised as a key development strategy for rural areas. It is now accepted that AT innovations can free rural women from drudgery and time-consuming labour. This saves them time and energy, which they could use to engage in self-development activities such as education and training in new skills. However, the available information indicates that rural women are not utilizing the available technologies to a significant extent. The purpose of this study therefore was to establish factors affecting adoption of appropriate technologies by rural women. The study was designed to carry out a survey on a target group of rural women. A random sample was obtained from an accessible population of rural women in Meru and Kiambu districts where AT programmes were initiated. A total of 160 respondents who included users and non-users of AT were selected. The research instruments used to collect data included an observation checklist and a structured interview schedule. The data collected and used in this study included types and conditions of technologies adopted by respondents; demographic and social economic status of respondents; experiences in adoption process and utilization of devices, and perception of rural women on AT. The data were analysed through descriptive statistics-tests analysis chi-square test of association and Pearson’s Product moment correlation analysis. The findings show that the majority of respondents were married, middle aged women, who had attained a primary school level of education. They were in occupations that combined subsistence farming with housework and other related activititesConsequenly, the women carried daily heavy workloads of more than 15 hours. The factors that were found to facilitate adoption of AT included long duration of stay in the community, age, family income, salaried occupation encouragement by husbands and participation in women groups. On the other hand, the major constrains inhibiting adoptions of technologies were found to be financial problems; lack of awareness. Others include inappropriate policies in women projects, biases on women, lack of access to technologies, complexity in implementation and adaptability of technologies. Some of the technologies adopted by women in Meru and Kiambu were improved stoves, dish lacks, cement water jars, ventilated improved latrines, bole holes, solar energy, biogas, soakpits and charcoal coolers. Most of the women kept their devices in good conditions and utilized them well. Home Economics Extension played a significant role in the adoption processor AT devices by ruler women and in the utilization of devices. Statistically significant relationship existed between adoption of technologies and some key variables. These key variables were age of the users and the level of education; users' rating of rural women's work load and age; types of occupation and belief that lack of formal education hinders adoption technologies by rural women; users' level of education and the study of home science in schools. A comparison of users and non-users in respect of AT devies used showed that the users had resided in communyties for longer durations of time, were older and were in more regularly salaried occupation than non users. Generaly Kiambu women found to be significantly older than those in Meru and also they had stayed in their communities for a significantly longer period of time than the womenin Meru. Kiambu women users and Meru users differed significantly in terms of types of technologies adopted, duration of time technologies were used and maintenance condition of devices. Meru users kept hteir devices iin significantly better condition than Kiambu women. However, users in both districts were similar in the way they perceived the need for technologies.
Developing Instructional Materials that Address Challenges Facing Teachers in Secondary School Chemistry Investigative Practical Work; A Case of Kajiado County, Kenya
(Kenyatta University, 2014-10-08) Ituma, Monica Gakii
Secondary school Chemistry teachers use various instructional materials to guide the teaching of Chemistry practical work. Many of them however face some challenges in their endeavor to implement investigative type of practical work in their classes. They often require instructional materials that would support the implementation of learner-centred investigative practical work. The purpose of this study was to develop a model of exemplary instructional materials that can support secondary school Chemistry teachers in engaging learners in investigative practical work by addressing the challenges they faced when teaching practical work. The study is based on constructivist theory of learning which proposes meaningful learning by construction of knowledge gained through context-rich, experience-based activities. The study used Design Based Research (DBR) design methods for analysis, design, development and evaluation of the exemplary materials. DBR uses iterative design to develop workable interventions in educational practice. Forty two (42) government secondary schools in Kajiado County formed the target population. Baseline data revealed that teachers needed support in content knowledge, scientific practices, scientific literacy practices and teaching strategies for participation practices and for assessment of learning. Instructional materials for a total of six lessons were designed as derived from the Form one Chemistry topic on acids, bases and indicators. The first prototype of materials designed was appraised by 47 chemistry teachers in pre-service training, three experienced Chemistry teachers and two Science Education experts from the University. The feedback was used to redesign and refine the materials producing the second level prototype which was tried out by three other teachers with their Form one students. Feedback gathered from the try-out was used to re-design and refine the instructional materials leading to production of a third level prototype that was used by five teachers as the practicality and effectiveness of the materials was evaluated. Lesson observation, teacher‘s logbook, teacher interviews, learner questionnaire and concept maps were used to determine the practicality and effectiveness of the materials in a classroom set-up. The instructional materials were found to contain guidelines that teachers could use in guiding learners through investigative practical work. Teachers indicated that their objectives were achieved and learners were motivated to learn chemistry and were able to understand the concepts. The feedback was used to redesign the materials thus developing the final model of exemplary materials. The structure of such materials was detailed and a model for the development of instructional materials for investigative practical work referred to as Secondary Chemistry Investigative Practical Work (SCIPW) proposed. Chemistry teachers can use the model to develop materials that guide them through use of learner centred strategies in practical work. Developers of instructional materials should also use such experimental structures that support investigative practical work.
Simulation and Optimisation of a Drying Model for a Forced Convection Grain Dryer
(Kenyatta University, 2018) Osodo, Booker Onyango
Forced convection grain dryers are more efficient and achieve greater drying rates than natural convection dryers. However, it is necessary to provide an appropriate solar air heater in order to achieve the required drying air temperature. Well sized fan and drying cabinet, as well as an optimal combination of air velocity, temperature and grain layer thickness are also essential for improved performance of such a dryer. In order to predict variation of moisture content with time during the drying process, it is necessary to have an appropriate drying model. In this study carried out at Njoro, Nakuru County in Kenya, an experimental grain dryer was sized, fabricated and its performance investigated under different drying conditions. Simulation of air flow within an initial model of the dryer was done and the results used to size the fan and drying cabinet. The effect of air velocity, grain layer thickness, number of trays and temperature on drying efficiency (ratio of energy used in removing moisture to sum of energy lost by drying air and that used for running fan) and moisture removal rate (ratio of mass of moisture removed to mass of wet grain per unit time) was investigated. The Taguchi approach was used to determine the optimal combination of drying air velocity, temperature and grain layer thickness that could be used to ensure greatest drying efficiency and Moisture Removal Rate (MRR). Analysis of Variance (ANOVA) and Least Square Differences (LSD) tests were used to determine whether change of air velocity and grain layer thicknesses significantly affected drying efficiency as well as MRR. The best fitting drying model for drying maize grain was selected and subsequently used to develop a computer simulation model for predicting drying time. On the basis of simulation results, number of trays and mass of grain to be dried per batch, the experimental grain dryer developed was of dimensions 0.5 m x 0.5 m x 1.0 m and was equipped with a 0.039 kW centrifugal fan. MRR was found to decrease with increase in grain layer thickness as long as air velocity was kept constant. For example, at 0.41 m/s air velocity, as grain layer thickness increased from 0.02 to 0.08 m, MRR decreased from 0.061 to 0.022 kg moisture / (kg wet grain. hour). Drying efficiency decreased with increase in drying air temperature where-as MRR increased with rise in air temperature as long as air velocity and layer thickness remained constant. For an air velocity of 0.41 m/s and 0.04 m grain layer thickness, drying efficiency was 23.5% at 40 °C and reduced to 10.1 % at 55 °C. On the other hand, MRR increased from 0.045 to 0.058 kg moisture / (kg wet grain. hour) over the same temperature range. It was found that when drying a given grain layer thickness, use of two trays did not significantly improve MRR as compared to the use of one. As a result of the optimisation process, it was also determined that when drying was done under laboratory conditions, a combination of 0.41 m/s air velocity, 45 °C air temperature and 0.02m layer thickness resulted in greatest MRR and drying efficiency. The drying model that best describes the drying curve was found to be the Midilli model. The optimal drying parameters, if applied by the user of the dryer, will result in optimal drying rate and drying efficiency, and this in turn will lead to reduced post-harvest grain loss. The computer simulation model developed will enable the farmer to plan drying schedules. Application of simulation to size the fan and dryer cabinet should be emulated by those who seek to size dryers. It is recommended that further study be carried out to determine the effect of grain porosity on dryer performance. Investigations should also be done to find ways of utilizing the warm exhaust air from the dryer.
Design Optimization of Municipal Solid Waste Incinerators Using Mathematical Modeling and Computer Simulation
(Kenyatta University, 2020-07) Sarakikya, Halidini Hatibu
Environmental pollution caused by the Municipal Solid Waste (MSW) incinerators has raised concerns about the quality of incinerators and the incineration process in Tanzania. Engineers and Scientists in general have appreciated that the installation of functional incinerators will increase the incineration process efficiency. Among the methods to achieve this is the application of mathematical modeling for the incineration process. Literatures have showed that related study in incinerators, incineration process and mathematical modeling in general has received little attention. The broad objective of this study was to optimize the design of municipal solid waste incinerators by using mathematical modeling and computer simulation. Computation Thermal Predictions (CTP) mathematical relations applying different types of incineration parameters including temperature, density, velocity and species concentration were formulated based on theories of incineration process with proper assumptions. The solution of the mathematical model developed was done and accomplished by computational fluid dynamics (CFD). Finite element method of numerical analysis was applied during the process to govern temperature and species concentration at different stack height of the model incinerator. Tests were performed on the physical model incinerator and data were analyzed after experiments. These data were applied in testing and verification of the mathematical models and provided the exact temperature and amount of flue gases which can be released from the stack without polluting the atmosphere. The results show that it is possible to forecast temperature and flue gases by the application of mathematical expressions. It also can be applied to develop more and accurate computational thermal predictions (CTP) model for the simulation of incineration process and experimentally regulate temperature and species concentration at different location of the incinerator. The results provided by the computational thermal predictions (CTP) were very close to those of experiments obtained from physical model. Therefore, there is an agreement between the empirical model and experiment as they show true trend of the incineration process.
An educational data mining model for promoting self-regulated learning on learning management systems
(Kenyatta University, 2023) Eric, Araka N.; Elizaphan Maina; Robert Oboko; Rhoda Gitonga
Online learning environments such as Learning Management Systems (LMS) utilized by institutions of higher learning to deliver open and distance education have different features that may not adequately provide individualized support to learners. Online learners experience inadequate instructor support as many students are enrolling in fully online or blended courses. Therefore, the success of online learning depends on the learner’s ability to take control of their learning process as defined by the Self-Regulated Learning (SRL) theory. Since online learners have no restricted time to go online and learn, the existing freedom requires students who can take charge and control their learning. Moreover, existing models for supporting SRL do not link SRL in educational psychology and SRL in LMS. In view of this, it is necessary to have LMS that support SRL by providing targeted interventions based on students’ online learning activities data. This study explored the use of Educational Data Mining (EDM) techniques to support students’ SRL skills on LMS. To accomplish this, the research applied a mixed-methods approach that involved three phases. First, qualitative and quantitative methods were used in problem identification. Qualitative research was used to explore the current methods used to measure and promote SRL in online learning environments. Quantitative research was used to conduct a pre-study to establish how LMS features are utilized in teaching and learning in higher institutions of learning in Kenya. Secondly, Integrated Design Process was used to design clustering and classification algorithms and integrated them into Moodle LMS to analyze students’ learning activities data. Finally, true experiment research was used to establish the effectiveness of the developed EDM model in promoting SRL through sampled University students who enrolled in data science with python course for 12 weeks. This study, therefore, contributes an EDM model that contains algorithms derived from learning theories as applied in self-regulated learning. Results indicate that students were able to progress from poor self-regulators cluster level to good self-regulators and exemplary self-regulators. The results reveal that students were able to evolve from one cluster to another over time as a result of receiving EDM interventions during the online course. The findings demonstrate that providing external support through the use of EDM prompts leads to the growth of students’ SRL skills. Moreover, this study reveals that it is possible to measure and promote SRL concurrently using EDM techniques. Future studies can focus on the evaluation of the EDM model under varied contextual conditions to examine the effect of SRL interventions on students’ SRL skills and academic performance.
Thermodynamic modelling and optimization of biomass gasification system for fischer-tropsch synthesis
(Kenyatta University, 2023) Kombe, Emmanuel Yeri; Nickson K. Lang’at; Paul M. Njogu
Biomass is a feasible route for producing transport fuels through Fischer-Tropsch synthesis (FTs). Production of transport fuels from biomass gasification output gas by FTs has become more attractive due to its ability to substitute fossil fuel in the energy market. Gasification technology is at the forefront of biomass conversion among other technologies due to its high flexibility in utilizing various biomass materials. In this work, a thermodynamic model of air gasification of rice husk in a downdraft gasifier was developed using Aspen Plus software and Engineering Equation Solver at various operating conditions. Experiments were conducted to validate the model. The influence of gasification temperature, equivalence ratio (ER), and moisture content (MC) on the composition of syngas, hydrogen to carbon monoxide ratio (H2/CO), and lower heating value of syngas was studied. Response surface methodology was applied to study the combined effects of the main operating parameters and thus determine the optimized zone of the operating conditions for Fischer-Tropsch synthesis. The R2 values of the generated regression models from the ANOVA tool were observed to be 98.47% for LHVSyngas, 98.93% for H2, 96.94% for CO and 89.91% for H2/CO molar ratio with corresponding Adj-R2 values of 98.28%, 98.80%, 96.56%, and 88.64%, respectively. This result indicates that the regression models determined the response variables with a high accuracy level. By using Response Surface Methodology (RSM), an optimization of the parameters was achieved. The RSM analysis results showed optimal conditions at gasification temperatures between 720 oC and 780 oC, ER in the range of 0.06 and 0.095, and MC in the range of 10% and 16%. The findings of this study reveal that a blend of simulation with advanced optimization tools can indeed achieve optimal operating conditions of a gasification system at a more refined precision. These analyses could form a basis for future practical development and implementation of biomass-based gasification systems through the selection of the best possible conditions in on-field plant operations
Hybrid machine learning model for comparative opinion mining in brand reputation monitoring
(Kenyatta University, 2024-11) Ondara, Bernard Omoi
Social media platforms like X platform (formerly Twitter) and online review websites like Amazon Reviews allow people to express their opinions about a brand’s products or services. To obtain competitive intelligence, brands can leverage this online user-generated content, through opinion mining, to extract useful insights to help them monitor their online reputation. Existing methods of brand reputation monitoring are mostly manual or automated to perform direct opinion mining with respect to a specific brand. In contrast, comparative opinions convey much more precise opinions about a specific brand relative to its competitors. Research in comparative opinion mining is rapidly gaining traction because of its extensive range of applications in areas such as brand reputation monitoring. Past studies utilizing machine-learning approaches have largely focused on applying single machine-learning models to perform direct opinion mining, targeting opinions about single entities. Results from the resultant tools are often misleading because they disregard opinions expressed towards other entities in comparative opinion data. Mentioning multiple entities in a comparative text potentially alters the polarity of opinions towards a target brand. Typically, existing models were built and tested using a limited number of comparative opinion labels and datasets, and were applied to a couple domains. Consequently, their reported performance may not be optimal in multi-label classification problems, comparative opinion mining, other application domains, and with larger datasets. Attempts at comparative opinion mining have largely focused on comparative sentence extraction using single machine learning models, thereby not leveraging the benefits of hybrid machine-learning models. In contrast, multi-label classification and exploitation of hybrid models consisting of machine learning models and/or deep learning models have shown performance improvements in model accuracy, transfer learning, data sparsityhandling, domain adaptation, robustness, and model generalization even on complex and huge datasets. Through systematic literature analysis, data analysis, empirical analysis, and statistical analysis methods, the researcher developed and validated a hybrid machine-learning model for comparative opinion mining using datasets from multiple domains. The model was applied to brand reputation monitoring for target brands as a proof of concept. The Multilayer Perceptron (MLP), which is a deep learning model, served as the base model because of its improved flexibility in feature extraction, minimization of prediction errors, and ease of integration with single models like Random Forest (RF) that served as the top-level model. The hybrid models outperformed the single models in accuracy and f1-score across multiple datasets, leveraging count vectors and trigram features. The lowest classification accuracy was 92.1%, while the highest was 93.0%. The MLP and RF hybrid model outperformed the other hybrid models and had a prediction efficiency of 0.1 milliseconds. The statistical tests show a significant difference between the performance (accuracy) of hybrid models and single models. Engaging three human experts in validating the hybrid model revealed that the hybrid models were generally more accurate and efficient than the single models. This is because hybrid models leverage the strengths while diminishing the weaknesses of single models. Therefore, hybrid models are more suitable for applications like brand reputation monitoring
K-means: critical analysis on the techniques used to determine the optimal value of k in high-dimensional datasets
(Kenyatta University, 2024-12) Gikera, Rufus Kinyua
Clustering is one of the main goals of exploratory data analysis. It has an extensive and wealthy history in a variety of fields. The methods used to perform clustering have been evolving over time. Among these methods, k-means is still the most popular clustering algorithm because of its ability to adapt to new examples and to scale up to large datasets. It is also easy to understand and implement and is computationally faster and more efficient compared to other algorithms. However, with k-means, selecting the correct k-hyperparameter, i.e. the number of clusters in a dataset, has a long standing challenge and has a significant effect on the clustering results. Although a number of k-hyperparameter tuning techniques in high-dimensional space clustering have been proposed, to help in the selection of the correct k-value, these techniques still face performance limitations in a variety of high dimensional datasets and dimensionality reduction methods. This makes the k-hyperparameter tuning problem intractable and an open research challenge. In light of this, this research firstly aims at investigating the existing k-hyperparameter tuning techniques in high dimensional space clustering through the literature review analysis. Secondly, an investigation on the dimensionality reduction methods used with the high dimensional spaces is also done via the same process. The results of the first two steps provide key findings and a conceptual framework that acts as the road map and the foundation for the subsequent empirical investigations in the third step. These investigations are guided by a comprehensive methodology based on mixed research methods for validation triangulation. Experiments are conducted on techniques that demonstrate methodological rigour and novelty, in a variety of datasets and dimensionality reduction methods. Empirical research design guides the process of conducting these experiments. The invaluable insights based on the results’ analysis of the experimental data, evinces the significance of the feature extraction process as a critical leverage point in the effective k-hyperparameter tuning process in high dimensions. This guides the implementation of a novel generalizable technique, through a multi-methodological system development methodology. This technique is then validated against the existing ones, using similar metrics, in order to evaluate its effectiveness. Statistical significance tests, using the ANOVA and the Kruskal-Wallis H statistic, demonstrate that the new technique is more superior. This is also evinced by the improved internal index scores, cluster visualizations as well as the presence of shorter whiskers and higher median (Q2) values in the whisker-box plots, in a variety of datasets. The new technique handles a variety of datasets, using an improved self-adapting autoencoder based on an unsupervised transfer learning strategy and a thoughtful configuration of both the architectural and training-related hyperparameter settings. This makes it effective in handling data sparsity and curse of dimensionality limitations inherent in high dimensional spaces. Future research aims at evaluating its efficacy in wider application domains, including a further comparative analysis of hybrid sets of best performing dimensionality reduction methods

Browse

Browsing PHD-School of Engineering And Technology by Issue Date

Results Per Page

Sort Options