/Behavior-Classification-of-Exposition-Visitors

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

Primary LanguageJupyter Notebook

Behavior-Classification-of-Exposition-Visitors

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

Data

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

Model Architecture

  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

Ablation Test & parameters setting

Final Rank

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

Others

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.