shadyAI4/UmojaHack-Africa-2022-Beginner-Track

# 🏆 UmojaHack Africa 2022 — Beginner Track | Air Quality Fault Detection > **Competition Result: 🥇 Ranked #17 out of all participants** --- ## 📜 Certificate of Achievement --- ## 📌 Competition Overview **Event:** UmojaHack Africa 2022 — Beginner Track **Challenge:** Air Quality Sensor Fault Detection **Task:** Predict `Offset_fault` — a multiclass classification label indicating sensor faults in air quality monitoring devices **Platform:** Zindi Air quality sensors sometimes develop faults due to environmental conditions. The goal of this challenge was to build a machine learning model that could detect the type of offset fault (if any) in PM2.5 sensors based on readings from two sensors alongside temperature and relative humidity data. --- ## 📂 Repository Structure ``` ├── Last_notebook.ipynb # Main competition notebook ├── README.md # This file └── certificate.png # Certificate of participation (add yours here) ``` --- ## 🔍 Approach & Methodology ### 1. Exploratory Data Analysis (EDA) - Visualized **missing values** across all columns using horizontal bar charts - Identified significant **outliers** in `Sensor1_PM2.5` and `Sensor2_PM2.5` using box plots - Analyzed **class imbalance** in the target variable `Offset_fault` ### 2. Data Preprocessing - Filled missing values in `Sensor1_PM2.5` and `Sensor2_PM2.5` with a fixed value of `200` (empirically found to work best) - Imputed missing values in `Temperature` and `Relative_Humidity` using the **median** (robust to outliers) - Outliers were intentionally retained as they were deemed informative for fault detection ### 3. Feature Engineering - **Sensor ratio:** `Sensor1_PM2.5 / Sensor2_PM2.5` — captures the relative difference between sensors - **Humidity level:** Binned `Relative_Humidity` into `low`, `medium`, and `high` categories - **Datetime features:** Extracted `month` and `hour` from the timestamp - Applied `pd.get_dummies()` for one-hot encoding of categorical …