From ratings to insights: machine learning for likert scale data interpretation

Choong, Andrew Kah Hoong (2026) From ratings to insights: machine learning for likert scale data interpretation. Final Year Project, UTAR.

PDF Download (2560Kb)

Abstract

Likert-scale questionnaires are a cornerstone of research, yet traditional analysis often overlooks nuanced inter-question relationships, while standard machine learning models frequently suffer from overfitting and lack the generalisability required to handle diverse survey datasets. This project developed and validated a web-based, interpretable framework that automates the analysis of user-uploaded survey data using an enhanced C4.5 decision tree algorithm. The system's primary technical contribution is a robust preprocessing pipeline that executes automated heuristic cleaning of identifier columns and integrates Chi-Squared feature selection to statistically filter irrelevant predictors. To ensure model simplicity, the implementation utilises Pessimistic Post-Pruning to algorithmically generalise the tree structure. A novel feature of the framework is the synthesis of quantitative decision rules with qualitative themes extracted from open-ended responses via Natural Language Processing (NLP), which are then interpreted by a Large Language Model (LLM) to provide narrative insights. System evaluation involved a real-world case study and a usability study with 40 respondents. Results from the Post-Study System Usability Questionnaire (PSSUQ) yielded an overall mean score of 1.84 (on a 1–7 scale), indicating exceptional user satisfaction. Expert validation further confirmed that the AI-generated interpretations successfully unveiled complex item relationships and mirrored manual research findings, effectively bridging the gap between complex survey data and actionable, explainable insights.

Actions (login required)