A Comparative Machine Learning Approach for Usability Evaluation of Localized Software in Afghanistan

Usability is one of the most important software quality attributes in determining user satisfaction, productivity and system acceptance. Afghanistan is a multilingual country, and usability problems with localized software applications may occur due to linguistic diversity, cultural differences, and scarce resources for evaluating the usability of software applications. These traditional usability assessment techniques, such as heuristic evaluation and laboratory testing, are difficult to implement in a capacity-constrained environment, as they require significant time, expertise and funds. The study suggests an automatic usability evaluation framework by machine learning techniques for localized software systems in Afghanistan. Data were gathered from ISO 9241-11 usability framework-based structured questionnaires from users of localized software applications. After collecting the responses, they were subjected to a pre-processing procedure involving cleaning, encoding and normalizing. Three supervised machine learning algorithms Decision Tree (DT), K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) were trained and tested with 80:20 ratio of training and test data. Model performance was evaluated using the accuracy, precision, recall, F1-score and confusion matrix analysis. Experimental results indicate that the highest accuracy of 91.4% is obtained by DT followed by SVM with 90.0% accuracy and KNN with 88.7%. The results show that the Decision Tree is a good model for balancing the prediction accuracy and interpretation of local software environments. The proposed framework has been designed to be both scalable and cost-effective, and can be used to complement existing usability evaluation methods, and can also be used to aid in software quality improvement efforts in multilingual and developing country settings.