Ofirmince Tulak Bamba, Nur Vadila, Sri Fitrawati, Vilna Wati Tedang, Asrawati

Naive Bayes dan Decision Tree: Studi Kasus Klasifikasi Kepuasan Pelanggan E-Commerce

Introduction

Naive bayes dan decision tree: studi kasus klasifikasi kepuasan pelanggan e-commerce. Evaluasi Naive Bayes & Decision Tree untuk klasifikasi kepuasan pelanggan e-commerce dari 5.000 ulasan Olist. Naive Bayes lebih akurat (80.70%) & efisien.

Abstract

Peningkatan belanja daring mendorong e-commerce untuk memahami kepuasan pelanggan melalui analisis ulasan otomatis. Studi ini mengevaluasi dan membandingkan kemampuan algoritma Naive Bayes dan Decision Tree dalam mengklasifikasikan tingkat kepuasan berdasarkan 5.000 ulasan dari platform Olist. Ulasan dikategorikan ke dalam tiga kelas, yaitu Tidak Puas, Netral, dan Puas. Pra-pemrosesan meliputi pembersihan data, ekstraksi fitur dengan TF-IDF, dan pembagian data 80% latih dan 20% uji. Evaluasi menggunakan metrik accuracy, precision, recall, F1-score, dan AUC. Naive Bayes menunjukkan akurasi lebih tinggi (80,70%) dibanding Decision Tree (73,90%) serta performa klasifikasi yang lebih stabil. Dengan demikian, Naive Bayes lebih efisien untuk klasifikasi kepuasan pelanggan berbasis teks pada ulasan e-commerce.

Review

This paper, "Naive Bayes dan Decision Tree: Studi Kasus Klasifikasi Kepuasan Pelanggan E-Commerce," presents a comparative analysis of Naive Bayes and Decision Tree algorithms for classifying customer satisfaction based on e-commerce reviews. The study utilizes a dataset of 5,000 Olist reviews, categorizing them into three classes: Dissatisfied, Neutral, and Satisfied. The primary objective is to determine which algorithm offers better performance for this text-based classification task, concluding that Naive Bayes significantly outperforms Decision Tree in terms of accuracy and stability. This research addresses a timely and relevant challenge for the e-commerce sector in automating the understanding of customer sentiment. Methodologically, the study is well-structured and employs standard practices for text classification. The pre-processing steps, including data cleaning and TF-IDF for feature extraction, are appropriate and commonly used. The dataset split of 80% for training and 20% for testing ensures a robust evaluation of model generalization. Furthermore, the comprehensive set of evaluation metrics—accuracy, precision, recall, F1-score, and AUC—provides a thorough and multi-faceted assessment of each algorithm's performance, enhancing the credibility of the findings. The use of a real-world e-commerce dataset like Olist reviews adds significant practical relevance to the study's conclusions. The core finding, demonstrating Naive Bayes' superior accuracy (80.70%) compared to Decision Tree (73.90%) and its greater stability, provides a clear recommendation for practitioners. This conclusion effectively addresses the study's objective, highlighting Naive Bayes as a more efficient choice for text-based customer satisfaction classification in e-commerce contexts. The paper offers valuable insights for businesses aiming to implement automated sentiment analysis. While the study effectively compares these two established algorithms, it also provides a strong foundation for future research to explore the integration of more sophisticated natural language processing techniques or deep learning models to further enhance classification performance and address potential complexities in customer reviews.

Full Text

Comments

You need to be logged in to post a comment.