Optimizing Footwear Recommendations using CNN-Based Visual Similarity and Health-Fit Filtering

Patil, Siddhesh Sunil (2025) Optimizing Footwear Recommendations using CNN-Based Visual Similarity and Health-Fit Filtering. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

The increased popularity of personalized and comfort-focused recommendations in fashion e-commerce has shown that the current visual recommendation systems have severe limitations due to prioritizing aesthetic appeal over practical functionality. The study describes a hybrid footwear recommendation system composed of three-phase framework combining CNN-based models and rule-based health-fit filtering. The experiment uses a curated version of the UT Zappos50K dataset, where four ergonomic properties, namely arch support, cushioning, breathability, and wide-feet compatibility, are characterized by means of metadata-based tagging and domain-specific heuristics. Phase 1 involves extracting visual embeddings that are based on pretrained CNN models such as ResNet50, VGG16, and EfficientNetB0 in order to provide unimodal similarity findings. In Phase 2, extension of the current retrieval setting to multimodal retrieval is performed by adding a filter of a binary health-fit tag to the embeddings. In Phase 3, health-fit attributes are predicted by a multi-label classification model that is trained based on footwear images. The evaluation was done based on Precision@k, macro F1-score, cosine similarity, and health-tag match rate. All models show good results, but out of all, EfficientNetB0 emerges as the most efficient. It performs better for comfort-related features like breathability and wide feet. Its capacity of producing compact and semantically dense representations makes it a good fit for hybrid rule-based filtering systems.

The given research introduces a usable, transparent, and ergonomically balanced recommendation framework that combines deep visual learning and medical-grade personalization in the domain of fashion technologies.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Niculescu, Hamilton UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence H Social Sciences > HF Commerce > Electronic Commerce H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Fashion Industry R Medicine > RA Public aspects of medicine > RA0421 Public health. Hygiene. Preventive Medicine > Personal Health and Hygiene
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	02 Jul 2026 14:43
Last Modified:	02 Jul 2026 14:43
URI:	https://norma.ncirl.ie/id/eprint/9446

Actions (login required)

View Item