馬拉松**博覽會參訪動線類別預測 (Behavior Classification of Exposition Visitors)

• The goal of project is to predict the best route among five routing options.
• Team member: Jack Liu, Barron Chang, Michael Chung, Leo

Dataset consists of "sniffer_loc", "created_time". We focus on using the information about "sniffer_loc".

1. RandomForest
   • Ensemble of multiple decision trees (tree-based)
2. LSTM/RNN
   • Traditional sequence prediction method
3. CatBoost
   • Gradient boosting tree-based method
4. Transformer-based (BERT & XLNet)
   • With the help of Multi-head self attention mechanism

Note: Under our attempts, we found that transformer-based models have better result and may have higher potential

• Our team reaches the top-3 on this leader board.
• The statistics on Aidea platform by 2022.6.13

Other details and discussion are stored in the .pdf file. Please find reference there if you're interested.

• Reference/note.md: the report.pdf and poster.pdf can help you understand more details.
• src/note.md: Summarize how our code works and the purposes of each files.