Multi-criteria evaluation and selection of web application design solutions

Abstract

This paper investigates the selection of design alternatives for web applications based on a combination of characteristics relating to ergonomics and cognitive interaction, accessibility and inclusivity, information architecture and visual design. The aim of the study is to formalise the selection procedure based on a linear combination of criteria and fuzzy preference relations, and to determine the conditions for consistency and robustness of the resulting recommendations. The research data consist of pairwise comparison matrices for three model alternatives regarding the quality of web application planning and design. Local priorities are listed using the method of normalised geometric means without intermediate rounding. For the base weight vector 0,64; 0,26; 0,10, the integral evaluations of the alternatives are 0,257991; 0,405855; 0,336153 respectively, and the degrees of non-dominance are 0,20; 1,.00; 0,52. Both procedures select the second alternative. Additionally, an analytical sensitivity analysis was carried out with a variable weight for ergonomics and a fixed weight ratio of 13:5 for the other two criteria. It was found that the transition threshold between the second and third alternatives for the linear combination is approximately 0,500826, whilst for preference ratios it is 0,307692. Within the interval between these values, the recommendations of the methods diverge. For independent relative perturbations of the weights followed by normalisation, the threshold amplitude for the linear model to retain a single winner was determined to be approximately 25,75 per cent. Furthermore, the robustness of the choice to independent multiplicative perturbations of local pairwise estimates was investigated. It has been established that, under the baseline weights, the second alternative retains its preference up to a limit disturbance amplitude of approximately 17,06 per cent; moreover, at the threshold of a decision change, the maximum consistency ratio of the matrices remains below the accepted threshold of 0,10. It is shown that the greater stability of the ranking based on preference ratios is linked to the use of the order of alternatives and does not in itself constitute evidence of higher estimation accuracy. The practical significance of the results lies in the ability to identify project decisions that depend on the aggregation method and to determine the need to refine the weights and initial assessments. The conclusions relate to a model example; verifying the quality of implemented interfaces requires a separate user study.