报告详情
PARE: Leakage-Aware Evaluation and Explainability for Lifestyle-Based Obesity Prediction
编号：99
访问权限：仅限参会人
更新：2026-11-10 16:45:47 浏览：10次
Online
报告开始：2026年11月21日 15:15（Asia/Kolkata）
报告时间：15min
所在会场：[S5] Data Science and Intelligent Analytics [S5-3] Data Science and Intelligent Analytics
暂无文件
摘要
Papers that use the Palechor–Manotas obesity dataset (2,111 records) regularly report seven-class accuracy above 90%, and the figure is usually taken to mean that lifestyle answers can predict obesity. We argue that two features of the data get in the way of that reading. The label is assigned from body mass index (BMI) thresholds, and BMI is computed from two of the predictors. Less well known, roughly three quarters of the records were produced by the Synthetic Minority Over-sampling Technique (SMOTE) before anyone split the data, so a test respondent may have synthetic descendants in the training set. We introduce PARE (Parent-Aware Real-respondent Evaluation), which tests only on original survey respondents and drops synthetic records traced back to them. We apply it to a three-branch model built from a graph attention network, a bidirectional long short-term memory (LSTM) network and a static encoder. Under the usual behaviour-only protocol the model scores 78.49 ± 1.12% over five seeds, and removing its graph branch costs 3.22 points (p = 0.008). Under PARE, every method we tried drifts toward the majority-class rate; the best balanced accuracy is 31.7%, compared with 15.2% for a constant guess, and the graph branch no longer helps. SHapley Additive exPlanations (SHAP) rankings move as well. Vegetable intake leads on the benchmark, but family history leads on real respondents, and the weight given to gender falls by about a quarter. Synthetic records should be generated after splitting, and results on this dataset should be reported under PARE.
关键词
Obesity classification, data leakage, synthetic data, SMOTE, evaluation protocol, graph attention network, bidirectional LSTM, SHAP, explainable AI.
报告人
Dipina Paul
PHD Scholar D Y Patil International University稿件作者
全部评论
重要日期
-
会议日期
11月20日
2026
至11月22日
2026
-
10月31日 2026
初稿截稿日期
-
11月07日 2026
提前注册日期
-
11月22日 2026
注册截止日期
主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论