Master Assignment
Customer purchase prediction through machine learning
Type: Master CS
Location: Utwente
Period: Oct, 2018 - Mar, 2018
Student: Seippel, H.S. (Hannah, student M-BIT)
Date final project: March 29, 2018
Supervisors:
Abstract:
Due to today’s transition from visiting physical stores to online shopping, predicting customer behavior in the context of e-commerce is gaining importance. It can in- crease customer satisfaction and sales, resulting in higher conversion rates and a competitive advantage, by facilitating a more personalized shopping process. By uti- lizing clickstream and supplementary customer data, models for predicting customer behavior can be built. This study analyzes machine learning models to predict a pur- chase, which is a relevant use case as applied by a large German clothing retailer. Next, to comparing models this study further gives insight into the performance dif- ferences of the models on sequential clickstream and the static customer data, by conducting a descriptive data analysis and separately training the models on the dif- ferent datasets. The results indicate that a Random Forest algorithm is best suited for the prediction task, showing the best performance results, reasonable latency, of- fering comprehensibility and a high robustness. Regarding the different data types, models trained on sequential session data outperformed models trained on the static customer data by far. The best results were obtained when combining both datasets.