INTEGRATED WEBSITE USABILITY EVALUATION MODEL USING FUZZY ANALYTICAL HIERARCHY PROCESS AND ARTIFICIAL NEURAL NETWORK

Abstract

Numerous websites in this contemporary time have been plagued with many usability issues which have hitherto made the websites not effective and efficient for users while searching for information. Consequently, different website usability evaluation models have been proposed to help in evaluating websites. However, most existing models are rather too ambiguous and not easy to use. Also, selecting and ranking websites based on usability with respect to numerous criteria have become a very important decision-making process among users. Additionally, there is no existing machine leaning model developed to classify websites usability based on user’s rating due to lack of usability ratings data. This thesis therefore proposes a new integrated usability evaluation model using Fuzzy Analytical Hierarchy Process (FAHP) with Artificial Neural Network (ANN). Five criteria of Speed (Spd), Navigation (Nav), Ease-of-use (Eou), Content (Con) and Aesthetic (Aes) obtained through factor extraction out of initial seven criteria proposed are used in the study. Six Nigerian universities websites with good webometrics ranking are used as alternatives. These are University of Ibadan (UI), Covenant University (CU), Obafemi Awolowo University (OAU), University of Nigeria Nsukka (UNN), University of Lagos (UNILAG) and Ahmadu Bello University (ABU) websites. Two sets of usability data were collected via google forms from 233 and 169 participants. Results from FAHP indicates that UI website has the highest global priority weight and hence is ranked as number one. This is followed by CU, OAU, UNILAG, UNN and ABU websites respectively. Also, final criteria weights obtained are 0.321Spd, 0.208Nav, 0.197Eou, 0.166Con and 0.108Aes respectively. This implies that the first and most important criteria to website users is speed. Weights obtained from FAHP model were preprocessed and used to train six machine learning algorithms which are Artificial Neural network (ANN), Random Forest (RF), Decision Tree (J48), Simple Logistic regression (SLOG), Bayesian Network (BaNET) and Logistic Model Tree (LMT). Results show that ANN has the best overall performance with accuracy (Acc) of 93.36% while RF, LMT, SLOG, J48 and BaNET have 90.12% Acc, 88.09% Acc, 88.18%Acc, 88.18% Acc and 83.63% Acc respectively. The FAHP model is further integrated with ANN to classify the user’s websites usability ratings. The ANN structure is 5-3-1 with logsig and trainbr as activation and transfer functions respectively. The best performance was obtained at learning rate (l) of 0.8, momentum (m) of 0.9 and threshold value(h) of 0.59. Further results obtained shows a precision (Pre), recall (Rec) and F-measure (Fme) values of 98.44%Pre and 95.45%Rec and 0.96Fme respectively. It is recommended that this integrated model, which can be used for users’ websites usability evaluation, ranking and prediction be adopted by IT practitioners and web developers.