對於電商產業來說，預測潛在顧客是重要的議題。本研究對於分析該議題有兩筆資料。第一筆資料來自於Kaggle 競賽『Acquire Value Shopper Challenge』，該筆資料含有顧客的購物紀錄及優惠券資訊，大小為22GB，主要目標為預測哪些顧客得到優惠券後仍會回購。第二筆資料來自台灣電商，該筆資料含有顧客的下單交易記錄以及商品相關資訊，主要目標為根據過往購物紀錄預測哪些顧客會再購買新商品。首先，我們會將數據流整合成新的資料矩陣，該矩陣中每列代表資料中每位顧客，特徵來自於整合其他數據流該顧客的過去交易資訊。在預測模型部分，我們使用邏輯斯迴歸(logistic regression)，梯度提升(gradient boosting) 以及 XGBoost 分類器。模型準則為AUC、預測準確率（accuracy）、查全率（recall）以及精確率（precision）。預測結果為電商公司提供有效資訊，使其廣告成本預算達目標。

Predicting potential customers is an important challenge for e-commerce companies. In this study, we analyze this problem from two data sets. The first dataset, based on the Kaggle competition called “Acquire Value Shopper Challenge”, contains all the given transactions with size over 22GB. The objective is to make predictions of customers who will remain loyal after a promotional period. The second dataset is from a local e-commerce company, which contains customer purchase history and product information. The objective is to make predictions for potential customers who will buy a new product given their past purchase information. We first transfer the data stream to a new data format where columns represent aggregated features and rows represent customers. The aspect of feature engineering is based on the aggregation of past customer transactions. We consider the prediction models including logistic regression, gradient boosting and XGBoost classifier. To evaluate the models, the area under the operating characteristic curve (AUC) and prediction accuracies are adopted. The results provide useful information for e-companies to keep their advertising budget in line with promotional and marketing goals.

論文審定書 i
論文公開授權書 ii
致謝 iii
摘要 iv
Abstract v
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Structure of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Machine Learning Model 1
2.1 Logistic regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.2 Gradient Boosting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.3 Extreme Gradient Boosting . . . . . . . . . . . . . . . . . . . . . . 2
3 Acquire Value Shopper Challenge 4
3.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2 Feature Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.3 Training set and Test set . . . . . . . . . . . . . . . . . . . . . . . . 7
3.4 Leave-one-offer-out method . . . . . . . . . . . . . . . . . . . . . 8
3.5 Prediction Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 Local E-commerce Company 10
4.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Feature Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 Training set and Test sets . . . . . . . . . . . . . . . . . . . . . . . 16
4.4 Prediction Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Conclusion 23

[1] Chen, T., and Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System. International Conference on Knowledge Discovery and Data Mining 6, 785-794
[2] Géron, A. (2017). Hands-on Machine Learning with Scikit-Learn & TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. O’Reilly Media.
[3] James, G., Witten, D., Hastie, T. and Tibshirani, R. (2013). An Introduction to Statistical Learning with Applications in R. Springer.
[4] Leo, B. (2001). Random forests, 45(1) : 5–32 Maching Learning Springer.
[5] Nikulin, V. (2015). On the method for data streams aggregation to predict shoppers loyalty. International Joint Conference on Neural Networks.
[6] Nikulin, V. (2016). Prediction of the shoppers loyalty with aggregated data streams. Journal of Artificial Intelligence and Soft Computing Research 6; 69-79.
[7] Acquire Valued Shoppers Challenge, part 1 - Competitions go through | Coursera. 2019. Coursera. https://www.coursera.org/lecture/competitive-data-science/acquire-valued-shoppers-challenge-part-1-nudnt.
[8] Acquire Valued Shoppers Challenge, part 2 - Competitions go through | Coursera. 2019. Coursera. https://www.coursera.org/lecture/competitive-data-science/acquire-valued-shoppers-challenge-part-2-b5i7N.
[9] Auduno. Kaggle code for Acquire Valued Shoppers Challenge. https://github.com/auduno/Kaggle-Acquire-Valued-Shoppers-Challenge.
[10] Sanket Doshi. Brief on Recommender Systems. https://towardsdatascience.com/brief-on-recommender-systems-b86a1068a4dd.
[11] XGBoost Documentation - xgboost 0.81 documentation. 2019. https://xgboost.readthedocs.io/en/latest/index.html.