Theses and Dissertation collection from the Faculty of Communication & Information Sciences


Recent Submissions

Now showing 1 - 8 of 8
  • Item
    Diabetes Mellitus (DM) is one of the most chronic and debilitating diseases in the modern society and represents not only a medical problem, but also a socio-economic problem. Computational Intelligence Techniques (CIT) have been successfully employed in diabetes disease diagnosis, risk evaluation, patient monitoring and prediction in medical field. Using single techniques in the diagnosis of diabetes has been comprehensively investigated showing some level of accuracy in Fuzzy Logic (FL) and Artificial Neural Network (ANN) for diagnosis of diabetes mellitus. Therefore, this study aimed at developing a Genetic Neuro Fuzzy Inferential System (GNFIS) for the diagnosis of diabetes. The objectives were to: (i) develop an enhanced hybrid system for diagnosis of diabetes mellitus using genetic algorithm; (ii) implement the enhanced system using Java programming language; (iii) evaluate the performance of the proposed system based on accuracy, sensitivity and specificity; and (iv) perform a comparison of the proposed system with two existing systems; Fuzzy Logic (FL) and Artificial Neural Network (ANN) for diagnosis of diabetes mellitus. The Neuro fuzzy inferential system was used to classify diabetes mellitus with genetic algorithm applied to obtain most relevant attributes from Pima Indian Diabetes Dataset (PIDD). Direct rating method was used for acquiring data used for the system. These data were presented in a series of objects to a domain expert who was requested to rate the membership function for the eight most significant attributes which also serves as input to the system. The ratings were aggregated for membership function calculation. The lowest was used as the minimum and highest as the maximum values. Triangular membership function was used because of its flexibility and fewer complexities when splitting values (low, medium and high), compared to other membership functions. The findings of the study were that: i. the enhanced hybrid system using genetic algorithm performed better during the diagnosis process for diabetes mellitus; ii. the enhanced system was implemented using Java Programming Language and it achieved a high level of accuracy; iii. the GNFIS gave a minimum diagnosis accuracy of 98.26%, maximum diagnosis accuracy of 99% andthe averageaccuracyof97.76%; sensitivity of 96% and specificity of 99% for the reduced dataset; and iv. results of comparison showed that GNFIS had a better performance with 99.34% accuracy on the whole dataset used when compared with FL and ANN with 96.14% and 95.14% respectively. The study concluded that the genetic algorithm is a good attributes reduction technique for Neuro Fuzzy Inferential System. The developed GNFIS performed efficiently and also outperformed the existing systems. Thus, the study recommended GNFIS as a good technique for the screening and diagnosis of DM.
  • Item
    (UNIVERSITY OF ILORIN, 2018-05) ALABI, IsiaqOludare
    Fraudulent activities with intent to defraud are pervasive in every system. The traditional statistical and long investigative auditors’ methods of fraud detection are becoming obsolete in dealing with the present trend of financial malpractices. The existing fraud detection methods are inadequate in terms of cost and time to detect emerging frauds. This study aimed at developing an enhanced Radial Basis Function (RBF) network to efficiently detect financial frauds. The specific objectives were to: (i) determine the input features relevant to fraud occurrences; (ii) create a series of RBF models; (iii) select an optimum model amongst the candidate models; (iv) evaluate the chosen model; and (v) compare the model with some existing ones. The methodology employed RBF algorithm with radially symmetric Gaussian activation functionsto classify previously unseen data features into their respective true categorical classes. A set of 20 input attributes, minimum hidden nodes and weight adjustments were observed for different runs until the network became divergent. The model was implemented in Rvectorized software packages and was trained 200 times with 1000 online German bank credit transactions. The developed RBF model for fraud detection was compared with three other models: Multi-layer perceptron back-propagation (MLP), Dynamic decay adjustment (DDA) and General nonlinear regression (GNLR). The findings of the study were that: i. twenty input features were ranked in order of their importance relative to the network’s output; 4 of the variables, namely, credit history, nationality, guarantor and job status were the most important in that order, while 5 variables: accounts type, telephone number, co-applicant’s guarantor and savings bond are the variables with negative contribution; ii. a total of 200 RBF base models were generated with randomly selected 850 credit transactions out of 1000 at ten different epochs and hidden nodes; iii. the RBF model for fraud detection was developed and of the 200 candidate base models, an optimum modelwas obtained at 600 iterations alongside 840 hidden nodes with a misclassification error rate of 6.9% and about 93% degree of accuracy. RBF models are of the form: y(x)= ∑_(i=1)^h▒〖ω_ij φ_i 〗 (x), where ω_ijis output weight, and φi(x) is the Gaussian activation function, 1/(1+e^(-φ_j (X)) ), with output y∈ {0,1}; iv. the chosen RBF model was evaluated and its performance yielded 7.18% misclassification error rate, 92.82% prediction accuracy, 89.71% sensitivity and receiver operating characteristic (ROC) of 98.6 %; and v. RBF had the highest average accuracy of 75.90% compared to DDA (75.04%), MLP (74.60%) and GNLR (74.53%). Further, RBF and DDA attained the highest accuracy score of (85.71%) and (79.22%) respectively, at iteration 4, while MLP and GNLR attained their highest accuracy at iterations 7 and 5 respectively. The study concluded that RBF network compared to other similar models trained faster, minimized the cost and time of fraud detection. The study recommended that RBF network is an efficient fraud detection model and its parameters must be adjusted closely to zero tolerance as a little threshold tolerance could imply a significant cost.
  • Item
    (UNIVERSITY OF ILORIN, 2018-05) ABISOYE, Opeyemi Aderiike
    The increase in the rate of malaria infections called for more rigorous research efforts. Evidences have shown that complexities involved in malaria forecasting required computational approach. Most existing conventional malaria forecasting models examine the dynamics of binary cases of asymptomatic of malaria parasite counts in the thick blood smear. Such existing models are plagued with the problem of over-fitting and prone to local minimum error due to large number of parameters to fix. Therefore, this study aimed at developing hybridized malaria prediction model for a multiclass nature of malaria parasite counts with effect of climatic conditions, given a non-linear malaria incidence cases. The objectives were to: (i) identify the factors that make an individual susceptible to malaria infection and the threats imposed; (ii) determine likelihood qualitative model of Artificial Neural Network (ANN) and Support Vector Machine (SVM); (iii) develop a thresholded hybridized SVM_ANN malaria model for better performance; and (iv) simulate the model to evaluate its performance compared with existing models using classification accuracy ( , sensitivity , specificity and mean square error (mse). Monthly survey of malarial incidence was collected from five randomly sampled health centers in Minna Metropolis, Niger State, Nigeria. Climatic data was also collected from the Nigerian Environmental and Climate Observation Programme (NECOP) Weather Station, Nigeria. These served as the model input variables. ANN with sigmoid transfer function for classification and SVM with radial basis function for feature selection were employed to predict the severity of multiclass malaria parasite counts. The findings of the study were that: i. the internal factors of frequency of human blood index, duration of sporogony, vector density, vector susceptibility, demography and external factors of temperature, rainfall, relative humidity vegetation, altitude, human behavioural factors and other environmental changes contribute to the possibilities of asymptomatic, symptomatic and climatic based threats; ii. SVM feature selection method produced optimal features for ANN classification; iii. hybridized SVM_ANN malaria predicting model was developed at an optimum threshold 0.60 and functionally expressed as SVM_ANN model = ϕ ( ) where ϕ (x) is the transfer function, = , ,xs, xc are support vectors with constraints of langrange multipliers αs, αc> 0 and αs, αc ∈αi and targets yi∈[0,1]; and iv. SVM_ANN hybridized malaria model achieved a better 98.91% , 100% , 98.68% and 0.14 mse compared with existing ANN and SVM model with 48.33%, 85.60% ; 60.61%, 84.06% ; 45.58%, 86.09% and 0.56 , 0.58 mse respectively. The study concluded that the prediction of multiclass symptomatic malaria infection with effects of climatic conditions using SVM_ANN hybridized model was established to be more efficient than existing models. Thus, this study recommended that the SVM_ANN hybridized model should be adopted by medical personnel and stakeholders to predict malaria incidence occurrences and its severity.
  • Item
    In face recognition system, several techniques have been proposed for extraction of facial features such as Local Binary Pattern, Gabor-filters, Elastic Bunch Graph Matching, Principal Component Analysis and Hidden Markov Models. Gabor-filters, among other feature extraction techniques, remain a powerful and useful tool in image processing. Its optimal functionality as feature extraction in face recognition is traceable to its biological importance and computational properties. In spite of all the distinctive characteristics of Gabor-filters, this technique suffers high feature dimensionality. This study therefore aimed at reducing the high dimensionality of Gabor features. The objectives were to: (i) extract facial features using Gabor-filters; (ii) optimize the Gabor features extracted with Ant Colony Optimization (ACO); (iii) perform facial image matching with the use of some selected distance classifiers; Chebysev, City-block, Mahalanobis and Euclidean; and (iv) evaluate the performance based on classification accuracy, classification time, sensitivity, specificity and error rate. The facial features was extracted using Gabor-filters with 5 scales and 8 orientations, then the extracted features were optimized by applying ACO on the Gabor features to obtain the optimal features. The optimized features was passed into selected distance classifiers. The performance evaluation of the proposed system was done using two face image datasets; Locally Acquired Face Image Database (LAFI) and Olivetti Research Laboratory Database (ORL). The findings of the study were that: (i) gabor feature vectors were obtained for face image representation; (ii) optimal features with relevant and discriminant information were produced; (iii) the optimized features performed efficiently with some selected distance classifiers; (iv) the best classification accuracy of 97.14% was obtained in Mahanolobis of image size (150x150) for LAFI Database, while classification accuracy of 95.71% was achieved in Mahanolobis (150x150), Euclidean (150x150), City-block (75x75, 100x100, 150x150) for ORL database; (v) reduced classification time of 0.42507secs was obtained in Mahanolobis (125x125) for LAFI Database and 0.40422secs was obtained in Mahalanobis (125x125) for ORL Database; (vi) the best sensitivity of 98.33% was obtained in Mahanolobis image size of (150x150), City-block (125x125) for LAFI, while the same percentage of 98.33% in Euclidean (150x150) for ORL; (vii) the best specificity of 90% was achieved in Mahanolobis image size of (150x150), Euclidean (75x75, 100x100, 125x125), Chebysev (75x75, 125x125) and City-block (75x75); and (viii) the best error rate of 2.86% was achieved in Mahanolobis of image size 150x150 for LAFI Database and 4.29% was obtained in Mahanolobis (150x150), Euclidean (150x150) and City-block (75x75, 100x100) for ORL database. The study concluded that the high dimensionality of Gabor features was well reduced and optimized by Ant Colony Optimization Algorithm. The performance of optimized Gabor features with the selected distance classifiers recorded better experimental results. Thus, the study recommended ACO as an effective feature optimization method for Gabor-features based face recognition system.
  • Item
    Office Space Allocation (OSA) is a major problem in higher institutions of learning. As a result of this problem, most of the demanding entities (staff) are wrongly allocated. The problem of OSA is considered to be Non-Polynomial (NP)-Hard combinatorial optimization problem which has been attended to by different researchers in the field of Artificial Intelligence (AI) and Operations Research (OR). Due to its combinatorial nature, several methods have been proposed, which include mathematical, heuristic and meta-heuristic methods. Considering the various methods available, meta-heuristic algorithms in their combinatorial forms need to be developed and tested for solving OSA in Nigeria Universities. Since the hybridization of the meta-heuristic algorithms considered in this research is not yet in existence, this study aimed at developing a hybrid meta-heuristic algorithms of Tabu search and Artificial Bee Colony in solving OSA problems using University of Ilorin as a case study. The objectives of the study were to; (i) formulate a mathematical objective function model for OSA problem and calculate penalty weight; (ii) adapt the algorithms to the problem of OSA; (iii) hybridize Artificial Bee Colony (ABC) algorithm with Tabu Search algorithm to solve OSA problem; and (iv) evaluate the algorithms using Halstead’s complexity measures. The research adopted a five-phase method. These phases included collection of dataset from the Faculty of Communication and Information Sciences, University of Ilorin, as a sample for mathematical modelling for solving OSA problem in terms of the objective function and the constraints. The methodology phases were adaptation of Artificial Bee Colony, Genetic and Tabu search meta-heuristic algorithms for the OSA problem, hybridization of ABC and Tabu Search algorithms to enhance the performance of the allocation, and a comparative study of the hybrid algorithms using halstead’s complexity measures. The findings of the study were that: i. the ABCgave lower penalty weight of 1678.3 when compared with 3885, 4036.6 and 1838.3 of hybrid, Tabu and genetic algorithms respectively; ii. when Tabu, ABC and genetic algorithms were adapted to the problem of OSA, the Tabu gave better result in term of time used. Tabu used 1231secs against 3114.8secs of ABC and 4256.3secs of genetic; iii. the hybrid algorithm of Tabu and ABC gave better result when compared with the three algorithms in the second finding in term of time used to solve the OSA problem. The hybrid used 616.62secs against 1231s, 3114.8s and 4256.3s of Tabu, ABC and genetic respectively; and iv. the halstead’s complexity measure such as program vocabulary, program length, program volume, program intelligence and program difficulty were used to compare the performance of all the algorithms and the hybrid algorithm gave the best result. The hybridized meta-heuristic algorithm and mathematical model developed was effective in solving the OSA problem and the use of population based algorithm enhanced the performance in allocating all entities to their respective offices. The hybrid algorithm also outperformed other existing algorithms considering the time used and the penalty weight. The study recommended the use of more hybridized algorithms in solving the problem of OSA in Nigeria Universities.
  • Item
    (UNIVERSITY OF ILORIN, 2019-09) OMOTOSHO, Oluwabusuyi Oladotun
    This study investigated the public libraries and peaceful coexistence among the residents in selected local governments in Kwara State. The study adopted descriptive survey design. The population comprised 27 librarians of Offa Township Library, Offa and Kwara State Library Board, Ilorin. A sample size of 27 librarians was used. Researcher used the entire population of librarians as sample size. Total enumeration sampling technique was adopted. Questionnaire was used as an instrument for data collection. Questionnaire was distributed to 27 librarians in the two selected public libraries, with 27copies of questionnaire returned. The response rate was 100%. Data collected was analysed and tabulated using simple frequency counts and percentage.The findings of the study revealed that there are various library services provided by public libraries in promoting peaceful coexistence among residents. Also, the findings of the study indicated that public libraries in Kwara State support government with various programmes that promote peaceful coexistence among residents.The findings of the study revealed that many prevailing factors such as; poor maintenance culture, management issues hindered peaceful coexistence among residents in Kwara State. Among the challenges identified are: inadequate resources and facilities, poor funding of public libraries by government and lack of interest in social issues by library staff. The study recommended that funding should be provided by government to public libraries in Kwara State. Also, the study recommended that qualified and dedicated personnel should be recruited into public libraries in order to ensure that effective service delivery towards peace promotion and conflict resolution is achieved.