Effective detection of purchasing intention for online shopping

Kang, Shu Yi (2023) Effective detection of purchasing intention for online shopping. Final Year Project, UTAR.

Preview

PDF Download (6Mb) | Preview

Abstract

The main issue with the below expectations in detecting purchasing intention is caused by the unbalanced data set and its overlapping class problem. To identify a sampling method that best improves the detection rate, this project performed four categories of sampling experiments, resulting in 2,011 experiments in total. To improve the detection results, a hybrid of undersampling and oversampling was applied to reduce and increase the size of the majority and minority classes of the unbalanced data set used in this project, respectively. Undersampling rates from 10% to 80%, and oversampling rates from 10% to 90% are used in combinations to achieve effective detections for the class "Buy", which is the minority in the data set. Random undersampling and five variants of Synthetic Minority Oversampling Techniques (SMOTE): Standard SMOTE, ADASYN, ANS, Borderline SMOTE, and SVM SMOTE, were utilised on the data set. Then, the resulting data sets were crossvalidated and tested with five classifiers: Decision Tree, Logistic Regression, Naïve Bayes, Random Forest and SVM. The result indicated that applying Random Forest with the random undersampling rate of 80% and oversampling rate (ANS) of 80% yielded the best recall in detecting the majority and minority classes overall.

Actions (login required)