Sensitivity Analysis with Fuzzy Sets
Identifying which inputs and rules drive fuzzy inference system robustness, applied to a noise-resilient movie recommendation system.
LUCID Lab, University of Nottingham, UK · Supervisor: Dr. Shaily Kabir · October 2024 – March 2025
Overview
Identifying which input variables and fuzzy rules determine the robustness and reliability of a fuzzy inference system, and evaluating how effectively non-singleton fuzzy sets (NSFLSs) handle uncertainty as dataset noise increases — applied here to a real-world movie recommendation setting.
Approach
1
System design
Designed a non-singleton fuzzy set (NSFLS)-based movie recommendation system, using the MovieLens dataset as the underlying rating data.
2
Noise modeling
Identified why review-based inputs used in movie recommendation are inherently uncertain, and constructed noisy versions of the dataset to test system robustness.
3
Comparative evaluation
Compared NSFLS-based recommendation performance against basic collaborative filtering and type-1 (singleton) fuzzy-based recommendation systems under varying noise levels.
Outcomes
Results show the NSFLS-based recommendation system consistently outperforming both collaborative filtering and singleton (type-1) fuzzy approaches as noise increases, demonstrating non-singleton fuzzy sets as a robust way to handle uncertain, noisy movie ratings.
This work is documented as a research white paper: “Towards Improved Movie Recommendations via Handling of Input Uncertainty”, co-authored with Tahmid Imtiaz, Dr. Shaily Kabir, and Dr. Christian Wagner.
Tools
FuzzyR Explainable AI