Abstract:
In the era of exponential data growth, the organization and labeling of data play
crucial roles. Unsupervised cluster analysis can be utilized to initially group the raw,
unlabeled data obtained from a large dataset. This thesis explores the impact of various
clustering algorithms, K-Means, DBSCAN, and Gaussian Mixture Models, on the
performance of a supervised classification model, specifically AlexNet. The primary
objective of the study is to evaluate the classification results on a subset of Places365
dataset after applying different clustering algorithms during the preprocessing phase.
Through a series of experiments, we demonstrate that the choice of clustering
algorithm significantly influences the performance of the classification model.