Checklist HAIO 2026 Summer Online Qualifier · B task
Production Line Quality Control
English title: Gyártósori Minőségellenőrzés
Predict the probability that a manufactured item is defective from sensor data, using 150 labelled and 12,850 unlabelled training samples.
The task
Each product on a (synthetic) production line is described by physical sensor readings (temperature, pressure, vibration, voltage, torque, humidity, rotational speed), maintenance data (tool wear, hours since last maintenance), five derived features (products and ratios of the above), eighteen further sensor values (Szenzor_01–Szenzor_18) and the production-line identifier (Vonal_A–Vonal_E). The target "Selejt" is 1 for a defective and 0 for an acceptable product.
Only 150 training samples are labelled; 12,850 further training samples are unlabelled, and the organisers point out that using the unlabelled data is worthwhile. The contestant outputs a defect probability for every test item. The starter notebook fits a baseline logistic regression on the labelled data only.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train_labeled.csv(150 rows: ID, 33 features, Selejt),train_unlabeled.csv(12,850 rows without Selejt) andtest.csv(2,000 rows). The repository folder B/adatok also containsbenchmark.csv(ID and Selejt probability columns) andsolution.csv(test labels).- You submit
- A CSV with header "ID,Selejt" giving a probability between 0 and 1 for every test ID, uploaded to Kaggle as <identifier>_B_<n>.csv, with the matching notebook uploaded through the DOCK interface.
- Scoring
- ROC-AUC.
- Rules
- Individual work only; no collaboration and no outside human help.
- The internet and any publicly available online tool, including AI-based tools, may be used; pretrained models are allowed unless the task description forbids them; no hardware restriction.
- At most 35 submissions per task per day; one submission per task must be selected for final scoring.
- The Kaggle team name must be the personal identifier sent by e-mail (e.g. HUN123); the notebooks must be uploaded (DOCK) within 15 minutes after the close.
- Everyone who submits valid solutions to both tasks advances to the national final; if that exceeds 80 contestants, the top 80 on the public leaderboard advance.
- Format
- Summer Online Qualifier (Nyári Online Előválogató) on Kaggle, 31 March 2026 09:00 – 2 April 2026 09:00 (48 hours); two tasks, final ranking by combining the rankings of both tasks.