This work addresses the challenge of classifying sophisticated bot traffic under the constraints of deployment in programmable network hardware. A novel dataset is presented that captures the behavioral differences between human and bot sessions in the context of e-commerce scalping. This dataset is generated using real browserbased human experiments and diverse automated bot clients. From the generated traffic, packet and flow features are extracted to train decision tree and random forest classifiers capable of identifying both bot and human traffic with high precision. To showcase the potential of in-network deployment, these models are converted into match-action tables and evaluated under the constraints of typical programmable hardware (e.g TCAM-based switches). While the training and test sets are limited in volume, results demonstrate that the models retain up to 89% accuracy with low false positive rates, generalizing well to unseen bot strategies. These findings highlight the viability of deploying bot detection for scalping events at line-rate using P4-based machine learning pipelines, offering a proof-of-concept for real-time mitigation of malicious scalping traffic at the network edge. Hardware deployment and evaluation of such classification models in real high-traffic events remains as future work.