Perbandingan Kinerja K-Nearest Neighbor (K-NN) dan Naive Bayes dalam Tata Kelola Teknologi Informasi Untuk menganalisa Tingkat Kepuasan Pengguna Aplikasi Shopee

Perbandingan Kinerja K-Nearest Neighbor (K-NN) dan Naive Bayes dalam Tata Kelola Teknologi Informasi Untuk menganalisa Tingkat Kepuasan Pengguna Aplikasi Shopee, K-NN, Naive Bayes, E-commerce, user satisfaction, IT governance, Shopee...

Author: Muhammad Wahyudi
Date: 2025
Keywords: K-NN, Naive Bayes, E-commerce, user satisfaction, IT governance, Shopee
Type: Jurnal
Category: penelitian

The rapid growth of the e-commerce industry in Indonesia compels companies to actively evaluate customer satisfaction through tingkat kepuasant analysis of app reviews. This study aims to compare the effectiveness of the K-Nearest Neighbors (K-NN) and Naive Bayes algorithms in assessing user satisfaction levels of the Shopee app as a component of IT governance. The research data comprises 2,000 samples of Shopee user reviews collected from Google Play Store and App Store, classified into five rating categories: excellent, good, fair, poor, and bad, based on textual content analysis. The preprocessing stage involved text cleaning (removing stopwords and punctuation), feature extraction using TF-IDF, and an 80:20 split of the dataset (training and testing). The analysis revealed significant differences between the two algorithms. K-NN achieved an accuracy of 54% with the optimal parameter *K=5*, while Naive Bayes demonstrated superior performance with 98?curacy. The low accuracy of K-NN is suspected to stem from its sensitivity to data imbalance and noise in text features, whereas Naive Bayes, with its probabilistic foundation, better handles the sparse characteristics of review data. These findings emphasize the criticality of selecting appropriate algorithms in IT governance for user satisfaction analysis. Naive Bayes is recommended as the optimal approach for text classification in e-commerce reviews, while K-NN requires refinement through techniques such as feature normalization or class imbalance handling. The study also highlights the need to integrate adaptive models, such as deep learning, to enhance accuracy in complex scenarios. For future research, expanding data scope or implementing multilingual analysis could serve as strategic steps to improve result generalization.

Files:
Tidak ada data !

Collections:
Digital Library UNPAB