Syifa Aurellia Rahma, Tiara Putri

Klasifikasi Konsumen Berdasarkan Loyalitas Belanja Online Menggunakan Algoritma Random Forest

Introduction

Klasifikasi konsumen berdasarkan loyalitas belanja online menggunakan algoritma random forest . Klasifikasikan loyalitas konsumen belanja online menggunakan Random Forest. Analisis transaksi e-commerce, identifikasi TotalPrice sebagai faktor dominan, dan dukung strategi retensi pelanggan dengan akurasi 91%.

Abstract

Pertumbuhan pesat aktivitas belanja online mendorong perusahaan e-commerce untuk memahami perilaku konsumen, khususnya dalam mengukur tingkat loyalitas pelanggan. Penelitian ini bertujuan untuk mengklasifikasikan loyalitas konsumen berdasarkan data transaksi penjualan menggunakan algoritma Random Forest. Dataset yang digunakan berasal dari platform Kaggle dan berisi informasi transaksi dari sebuah perusahaan e-commerce di Inggris. Data dianalisis melalui beberapa tahapan, termasuk pembersihan data, encoding atribut kategorikal, normalisasi data numerik, serta feature engineering untuk menghasilkan atribut seperti frekuensi belanja, total kuantitas, total pembelanjaan, dan durasi transaksi. Model Random Forest dipilih karena kemampuannya dalam menangani data besar dan tidak seimbang serta memberikan feature importance yang berguna. Hasil evaluasi menunjukkan bahwa model mampu mengklasifikasikan konsumen dengan akurasi sebesar 91%, meskipun performa pada kelas minoritas masih rendah akibat ketidakseimbangan distribusi data. Fitur TotalPrice terbukti menjadi faktor paling dominan dalam menentukan loyalitas pelanggan. Penelitian ini menunjukkan bahwa pendekatan klasifikasi berbasis data dapat digunakan secara efektif untuk mendukung strategi retensi pelanggan dalam industri e-commerce.

Review

This paper presents a timely and relevant study on classifying online consumer loyalty using the Random Forest algorithm. Given the rapid growth of e-commerce, understanding and predicting consumer loyalty is crucial for business sustainability. The research leverages a Kaggle dataset containing UK e-commerce transaction data, applying a comprehensive methodological pipeline that includes data cleaning, categorical encoding, numerical normalization, and insightful feature engineering. The core objective of classifying customer loyalty is well-defined, and the chosen approach aligns appropriately with the scale and nature of the data. A significant strength of this work lies in its robust methodology and commendable results. The selection of Random Forest is well-justified by its ability to handle large and imbalanced datasets, a common challenge in real-world transactional data. The authors effectively engineered valuable features such as shopping frequency, total quantity, total spending, and transaction duration, which significantly contribute to the model's predictive power. Achieving an impressive 91% classification accuracy demonstrates the model's overall effectiveness. Furthermore, the identification of 'TotalPrice' as the most dominant factor in determining customer loyalty provides a crucial insight that can directly inform targeted marketing and retention strategies within the e-commerce industry. While the overall accuracy is high, a notable area for improvement, acknowledged by the authors, is the model's lower performance on minority classes due to data imbalance. For loyalty classification, accurate identification of *all* classes (e.g., highly loyal vs. at-risk) is vital for actionable insights. Future iterations could explore advanced techniques such as oversampling (e.g., SMOTE), undersampling, or class weighting during model training to mitigate this issue and improve the recall/precision for minority groups. Additionally, while the paper mentions supporting retention strategies, a slightly deeper discussion on *how* these classifications can be translated into specific business actions or interventions for different loyalty segments would enhance its practical utility. Despite this, the research offers a valuable framework and provides solid evidence for the effectiveness of data-driven classification in e-commerce customer relationship management.

Full Text

Comments

You need to be logged in to post a comment.