A carregar...
Publicação
A/B Testing with a Cluster-based Control Group to Improve Restaurants’ Performance in Online Food Delivery Application
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.02 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
This research examines the integration of density-based clustering techniques into A/B
testing methodologies for online food delivery (OFD) platforms to address challenges posed
by heterogeneous restaurant performance data. The study employs DBSCAN clustering to
segment restaurants into distinct performance groups based on multiple operational metrics,
utilizing dimensionality reduction through UMAP to optimize cluster formation. Analysis
revealed eight well-defined clusters with silhouette scores exceeding 0.50 for restaurants
with lower variable1 values (categories 1-2) and five distinct clusters for those with higher
variable1 values (categories 3+). The clustering identified three primary discriminatory
variables (variable1, variable3, and variable4) that formed the basis for a systematic
classification framework with practical interpretability. While the clustering component
successfully created meaningful segments, implementation challenges in sampling
methodology, specifically non-randomized participant allocation, prevented conclusive
demonstration that cluster-based stratification produces superior experimental validity
compared to simple randomization. Statistical analysis using Welch's t-tests revealed
significant differences between experimental groups but also detected pre-existing baseline
differences that compromised internal validity. The research contributes methodological
insights on the conditions necessary for clustering to enhance experimental validity and
highlights the tension between rigorous experimental methodology and practical business
imperatives in industry settings. The findings suggest that clustering techniques can improve
A/B testing validity when properly integrated with randomization principles, though
implementation requires careful attention to fundamental experimental design
considerations. This work provides a foundation for future research on optimizing
experimental methodologies in heterogeneous business environments and developing
predictive applications that leverage cluster-based insights.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business Analytics
Palavras-chave
A/B Testing Clustering Algorithms Online Food Delivery Experimental Design Balanced Samples SDG 9 - Industry, innovation and infrastructure