Predicting customer satisfaction score in Brazilian e-commerce using machine learning techniques | |
| Author | Rumki, Mst Maria Rahaman |
| Call Number | AIT RSPR no.DSAI-25-03 |
| Subject(s) | Consumer satisfaction Machine learning Electronic commerce--Brazil--Data processing |
| Note | A research study submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Data Science and Artificial Intelligence |
| Publisher | Asian Institute of Technology |
| Abstract | The present study applies the methodology of the Brazilian e-commerce Olist data to examine how sophisticated machine learning (ML) and feed-forward neural network algorithms canbeusedtoforecastconsumersatisfaction rates. The four different customer satisfaction categories are "Low," "Average," "Good," and "Excellent," with "Excellent" evaluations being the most prevalent. Building upon prior research, we enhance the feature engineering process by introducing time delta, behavioral, temporal,and aggregated features such as delivery delay, approval duration, seller packing speed, purchase timing, product popularity, and seller reputation. Baseline model: product average (32% accuracy) is established to benchmark ML performance. To address data imbalance across classes, we apply both oversampling (BorderlineSMOTE) and undersampling (ClusterCentroids) strategies. A total of six ML models (RF,LR,K-NN,XGBoost,LightGBM and CatBoost) were evaluated,alongside a simple neural network like feed-forward neural network (FFNN) built in Keras. Randomized SearchCV was employed for hyperparameter tuning, optimizing models using macro F1-score due to class imbalance. With an average precision of 0.48, recall of 0.44, and F1-score of 0.45, the XGBoost model outperformed all other ML and Neural Network models, and BorderlineSMOTE significantly improved minority class prediction, according to experimental data. The five most important criteria influencing customer happiness, according to feature importance study, are product rating, rating stability, product popularity, delivery delay, and delivery duration. By demonstrating that enriched feature engineering, effective class imbalance handling, the application of advanced ML and FFNN techniques, and optimized hyperparameter tuning all contribute to notable improvements in predicting customer satisfaction scores, the study’s findings offer useful insights for enhancing the customer experience in ecommerce platforms. |
| Year | 2025 |
| Type | Research Study Project Report (RSPR) |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Data Science and Artificial Intelligence (DSAI) |
| Chairperson(s) | Chantri Polprasert |
| Examination Committee(s) | Mongkol Ekpanyapong;Chutiporn Anutariya |
| Scholarship Donor(s) | Tech Globe Scholarship |
| Degree | Research Studies Project Report (M. Eng.) - Asian Institute of Technology, 2025 |