# Classify Gases from Multimodal Information

*English translation by SOTA – AI Community of the Japanese original, shared under CC BY-NC-SA 3.0 IGO like the original. Organisers who would like this translation removed can email sota.ai.community@gmail.com.*

*Competition overview of the Kaggle competition "[Playground] JOAI Competition 2025": "A competition page equivalent to JOAI Competition 2025".*

## Overview

Note: this competition makes content equivalent to JOAI2025 available for practice. For the Private scores at the prize borders when the live contest ended, see [here](https://www.kaggle.com/competitions/playground-joai-competition-2025/discussion/640596).

In this competition, you will work on a classification task about gases using multimodal information: tables, images and text. Each data sample consists of two measurements from gas sensors, an image taken at the same time with an infrared camera, and a text describing the image. There are two kinds of gas, "Perfume" and "Smoke", and there are four labels that take into account the presence or absence of each (Perfume, Smoke, Mixed, NoGas). 

## Description

In this competition, you will build a model that classifies gases from multimodal information: tables, images and text. For details of the data, please see the "Data" tab.

The dataset for the competition was built from the following dataset.

>Narkhede, Parag; Walambe, Rahee ; Chandel, Pulkit; Mandaokar, Shruti; Kotecha, Ketan (2022), “MultimodalGasData: Multimodal Dataset for Gas Detection and Classification”, Mendeley Data, V2, doi: 10.17632/zkwgkjkjn9.2

## Evaluation

Submissions are evaluated with the F1 score between the predictions and the ground truth.

```
from sklearn.metrics import f1_score

y_true = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]
y_pred = [0, 1, 1, 1, 1, 0, 0, 0, 1, 1]

print(f1_score(y_true, y_pred, average="weighted"))
```

**What is the F1 score?**

It is a metric that combines the following two metrics, precision and recall.

  1. **Precision**  
     Of the predictions made, the proportion that were actually correct.  
     > Example: you predict "cat!" 10 times and are right 8 times → precision 8/10 = 0.8
  2. **Recall**  
     How many of the true cases were found without missing any.  
     > Example: there were actually 12 cats, but you found only 8 of them → recall 6/12 = 0.5
  3. **F1 score**  
     The "harmonic mean" of precision and recall.
    $$ F1 = 2 \times \frac{\text{Precision} \times \text{Recall}}{\text{Precision} + \text{Recall}} $$

*(Translator's note: the recall example is reproduced as in the original, which says 8 cats but computes 6/12.)*


## Submission File

For each index in the evaluation set, you must predict the gas (Gas) label. Create the file in the following format, including the header.

    index,Gas
    0,NoGas
    1,NoGas
    2,NoGas
    etc.

## Timeline

- 25 April 2025, opening ceremony – Start
- 2 May 2025, 23:59 – End
- 7 May 2025, 23:59 – Deadline for submitting the solution write-up (free format) and the source code (top-ranked participants in the selection track only)

All times are Japan Standard Time (JST) unless otherwise stated. The organisers reserve the right to update the competition schedule if they consider it necessary.