README.md

---
title: Weather & Aviation Mission Feasibility Analysis
task_categories:
  - tabular-classification
language:
  - en
tags:
  - weather
  - aviation
  - EDA
  - mission-feasibility
  - helicopter
size_categories:
  - 10K<n<100K
---

# Weather & Mission Feasibility Analysis

**EDA Assignment · March 2026**

> A decade of atmospheric data analyzed to determine the environmental thresholds that dictate helicopter mission suitability — bridging meteorology and aviation safety.

# Video Presentation

<video src="https://www.loom.com/share/6f05f5d4e5bb4636a826eff9951c2eb3" controls="controls" style="max-width: 720px;"></video>
<video src="https://huggingface.co/datasets/MichaelYitzchak/Weather_Conditions/resolve/main/weather_analysis.mp4" controls="controls" style="max-width: 720px;"></video>
The videos are in the required format but do not always work so Here is a backup link if the videos do not work
https://huggingface.co/datasets/MichaelYitzchak/Weather_Conditions/resolve/main/%3Cvideo%20src%3D%22...%22%20controls%3D%22controls%22%20style%3D%22max-width%3A%20720px%3B%22%3E%3C/video%3E.mp4


| Dataset | Period | Rows | Features | Questions |
|---|---|---|---|---|
| Szeged Weather · Kaggle | 2006–2016 | 96,453 | 12 | 5 |

---

#Main Objective

**To analyze to what extent atmospheric variables — humidity, altimeter pressure, and wind speed — can evaluate the suitability of environmental conditions for helicopter flight.**

This project focuses exclusively on hours with active weather conditions. There are no clear or sunny baseline days in the dataset — every row represents a moment where weather played a role. The engineered `Mission_Status` variable defines, based on standard aviation thresholds, whether conditions on any given hour were safe enough to fly.

The personal motivation: as a helicopter operator serving in the reserves, multiple missions have been cancelled mid-flight due to weather. A data-driven pre-flight weather assessment framework could save time, manpower, and allow missions to be replanned rather than abandoned.

### Mission Status Thresholds

| Status | Condition |
|---|---|
|  **Operational** | Visibility > 4 km **AND** Wind Speed < 45 km/h |
|  **Non-Operational** | Any breach of the above thresholds |

---

##  Dataset Description

**Source:** [Weather in Szeged 2006–2016](https://www.kaggle.com/datasets/muthuj7/weather-dataset) — Kaggle  
**Size:** 96,453 rows · 12 features

### Features

**Numeric:** Wind Speed (km/h), Visibility (km), Temperature (C), Apparent Temperature (C), Humidity, Pressure (millibars), Wind Bearing (degrees)

**Categorical:** Summary, Precip Type, Formatted Date

**Engineered:** `Mission_Status` — Operational / Non-Operational

---

##  Data Cleaning

| Step | Action | Result |
|---|---|---|
| 1 | Date parsing | Converted to UTC datetime |
| 2 | Duplicate removal | 24 rows removed → 96,429 remaining |
| 3 | Pressure sensor failures | 1,288 zeros (1.34%) replaced with column median |
| 4 | Missing Precipitation Type | 517 rows (0.54%) dropped |
| 5 | Redundant columns | `Loud Cover` (zero variance) and `Daily Summary` dropped |
| 6 | Humidity validity | All values confirmed in [0, 1] — no action needed |
| 7 | Visibility ceiling | All values ≤ 16.1 km confirmed — no action needed |
| 8 | Fog/Visibility contradiction | 1,939 rows relabeled from "Foggy" → "Partly Cloudy" where visibility > 2.5 km |

> **Note:** All numeric columns had zero NaN values (confirmed via `df.info()`). No general median-fill was required — only the targeted pressure zero replacement above.

---

##  Outlier Detection & Handling

Distribution shape was checked before selecting any outlier method — IQR assumes symmetry and fails on skewed variables.

![Distribution Shape Check](distribution_check.png)

| Column Group | Method | Decision |
|---|---|---|
| Temperature, Apparent Temp, Pressure | IQR (symmetric) | **Keep** — real weather extremes |
| Wind Speed, Visibility, Humidity | 1st–99th percentile (skewed) | **Keep** — operationally critical events |
| Wind Bearing | None (circular variable) | **Keep** — 0° and 360° are the same direction; IQR/percentiles fail on circular data |

**Overall decision: Keep all outliers.** In aviation weather analysis, extreme readings are not errors — they are the most operationally significant events in the dataset.

---

##  Class Balance

| Class | Count | Percentage |
|---|---|---|
|  Operational | 8,998 | 90.0% |
|  Non-Operational | 1,002 | 10.0% |

**Imbalance ratio: 8.98:1** — A naive model predicting "Operational" every time would score 90% accuracy while being completely useless. Future models should use **F1-score or precision-recall**, not raw accuracy.

---

##  Correlation Analysis

![Correlation Heatmap](correlation_heatmap.png)

**Key correlations:**
- Humidity ↔ Visibility: **–0.37** (negative — high humidity predicts low visibility)
- Wind Speed ↔ Visibility: **+0.10** (near-zero — they are **independent** risk factors)
- Temperature ↔ Apparent Temperature: **+0.99** (expected — nearly identical variables)

---

##  Research Questions & Findings

---

### Q1 · The Fog Trap

**Research Question:** To what extent does humidity serve as an indicator for non-operational mission status based on visibility?

![Humidity vs Visibility](q1_humidity_visibility.png)
*Left: Scatter plot of Humidity vs. Visibility with 4 km safety threshold · Right: KDE density of humidity by Mission Status*

**Finding:** A "Fog Trap" was identified at **90% humidity**. As humidity approaches 100%, visibility consistently crashes below the 4 km safety threshold — making humidity a primary **leading indicator** of mission grounding before conditions become critical.

---

### Q2 · Atmospheric Pressure Stability

**Research Question:** Does atmospheric pressure stability vary between operational and non-operational mission statuses?

![Pressure Stability](q2_pressure_stability.png)
*Left: Box plot — pressure spread by mission status · Right: Overlaid pressure distributions by mission status*

**Finding:** Non-operational missions show **higher pressure volatility** (std: 9.78 vs 7.29 for operational). Rapid pressure drops correspond to storm systems; high-pressure outliers indicate anticyclonic conditions. Pressure instability is a meaningful — if subtle — indicator of unsafe conditions.

---

### Q3 · The Wind Speed Ceiling

**Research Question:** To what extent can extreme wind speed variations predict a transition from operational to non-operational mission status?

![Wind Speed Violin](q3_wind_violin.png) ![Wind Speed Histogram](q3_wind_histogram.png)
*Left: Violin plot — wind speed by mission status · Right: Full dataset wind speed distribution*

**Finding:** A hard ceiling exists at **45 km/h** — beyond this, 100% of missions are Non-Operational. However, most non-operational events occur below 20 km/h, confirming **visibility is the dominant grounding factor**, not wind. Wind and visibility are independent (correlation: +0.10) — both must be monitored separately.

---

### Q4 · Categorical Weather Hazards

**Research Question:** To what extent does categorical weather data (Precipitation Type) serve as a reliable indicator for non-operational mission status?

![Precipitation Risk](q4_precipitation.png)
*Left: Mission count by precipitation type · Right: Proportional grounding risk (%) per type*

**Finding:** While **Rain** is the most frequent hazard in absolute numbers, **Snow** carries a ~40% higher proportional grounding risk. Most common Non-Operational weather: Foggy. Most common Operational weather: Partly Cloudy. Snow is a disproportionate predictor despite lower frequency.

---

### Q5 · Seasonal Risk Patterns

**Research Question:** Does the risk of mission grounding follow a predictable seasonal pattern throughout the year?

![Seasonal Risk](q5_seasonal.png)
*Left: Monthly grounding risk % (line chart) · Right: Normalized stacked bar — Operational vs. Non-Operational ratio by month*

**Finding:** The data reveals a dramatic **U-shaped seasonal curve**. January grounding rate: ~27%. December: highest risk. July: lowest risk. Winter missions are roughly **30× more likely** to be Non-Operational than summer missions — enabling data-driven strategic fleet planning.

---

##  Key Insights

| Insight | Detail |
|---|---|
|  **Wind & Visibility are independent** | Near-zero correlation — either alone can ground a mission. Monitor both separately. |
|  **Humidity is an early warning signal** | Above 90%, visibility reliably drops below 4 km. Predict groundings before they happen. |
|  **Snow is disproportionately dangerous** | ~40% higher grounding risk than rain despite lower frequency. |
|  **Seasonality is predictable** | December/January are ~30× riskier than July. Plan resources accordingly. |

---

##  Limitations

- `Mission_Status` is **rule-based**, not naturally observed — engineered from predefined thresholds that may not capture all real-world operational nuance.
- Results reflect **associations**, not causal relationships.
- The **90/10 class imbalance** means future classifiers need careful metric selection (F1, precision-recall) and should consider resampling techniques.
- Some variables may be **indirectly related** through confounders — seasonality drives both temperature and humidity, which in turn influence visibility.

---

##  Conclusion

This analysis demonstrates that aviation mission feasibility is strongly governed by environmental conditions. Through systematic data cleaning, distribution-aware outlier analysis, feature engineering, and five focused research questions, meaningful and actionable insights were derived from raw weather data.

Wind speed and visibility define hard safety boundaries; humidity serves as an early warning signal; pressure instability flags turbulent conditions; snow is a disproportionate categorical risk; and seasonality enables proactive operational planning. Together, these findings form a robust data-driven foundation for aviation safety decision-making.

---

*Weather in Szeged 2006–2016 · Kaggle Dataset · Python · Pandas · Matplotlib · Seaborn*