This study investigates the feasibility of predicting the outcome of e-commerce dialogues using machine learning models. Syntactic, semantic, and conversational features are extracted from dialogue utterances, including sentiment dynamics, intent diversity, and dialogue structure, and aggregated at the dialogue level. A Random Forest classifier and a Long Short-Term Memory (LSTM) network are applied to classify dialogues as successful (leading to a purchase) or unsuccessful. Results indicate that static Random Forests struggle to capture sequential patterns inherent in dialogue, achieving limited accuracy, while optimized LSTMs effectively leverage the sequential dynamics of conversations, providing robust and balanced classification. Additionally, a preliminary real-time LSTM approach shows the potential for turn-by-turn behavioral prediction, although predicted probabilities remain close, highlighting areas for future improvement. This work contributes to understanding dialogue-based user behavior modeling and its applications in adaptive e-commerce systems.