We’ve all had those days where we step outside and immediately regret our outfit choice—maybe you wore a hoodie on a hot day or forgot a raincoat when it started pouring. That’s exactly what this AI agent is here to solve!
This smart assistant doesn’t just predict the weather—it also tells you what to wear so you can stay comfortable and prepared, no matter what the forecast brings.
| Component | How It Works in Our AI |
|---|---|
| Performance Measure | The agent is judged by how accurately it predicts the weather and how well it recommends clothing based on that prediction. If the weather forecast is right but the clothing suggestion is off (e.g., recommending shorts on a freezing day), that’s a failure. The goal is to keep users comfortable while ensuring accurate weather forecasts. |
| Environment | The AI operates using Seattle’s historical weather data (2012-2015) and would ideally integrate with real-time weather updates in the future. It considers things like rain, temperature, and wind speed to make both forecasts and outfit suggestions. |
| Actuators (Outputs) | Instead of just predicting “Rain” or “Sunny,” the AI recommends what you should wear based on the weather, including:
|
| Sensors (Inputs) | The model takes in key weather factors to make predictions:
|
The whole idea behind this utility-based agent is simple: knowing the weather isn’t enough—people need to know how to dress for it. Instead of just giving you a generic weather forecast, this AI helps you make smarter clothing choices, so you’re always prepared.
We built this system using a Random Forest Classifier, carefully optimizing it to accurately predict both common and rare weather conditions (like fog or drizzle, which are often misclassified in other models). With this approach, the AI is better at handling real-world weather variations and giving more reliable recommendations.
So next time you’re wondering, “Should I bring a jacket?”—just ask the AI. It’s got you covered!
Ever stepped outside and instantly regretted your outfit choice? Maybe you didn’t check the forecast, or maybe the forecast wasn’t detailed enough to tell you what to actually wear. That’s exactly what this AI is built to fix.
Instead of just saying, "It’s going to rain today," our model helps you decide:
- Should you wear a jacket? If so, what kind?
- Is it warm enough for shorts, or should you stick with pants?
- Will the wind make it feel colder than it actually is?
To do this, we trained the AI on Seattle’s historical weather data (2012-2015), which includes everything from temperature to wind speed.
Each row in the dataset represents one day of weather, giving the AI key details to make smart predictions.
| Feature | What It Represents | Why It's Important |
|---|---|---|
| Date | The specific day (YYYY-MM-DD). | Not directly used, but seasonal patterns matter. |
| Precipitation | Rain or snowfall that day (mm). | Helps decide if a raincoat or snow jacket is needed. |
| Temp Max | The highest temperature that day (°C). | Determines if short vs. long sleeves are best. |
| Temp Min | The lowest temperature that day (°C). | Helps pick between shorts or pants. |
| Wind Speed | How strong the wind was (m/s). | Determines if a windbreaker is necessary. |
| Weather | The overall condition (rain, sun, drizzle, fog, snow). | This is what the AI predicts! |
Below is a diagram of the general overview of what each variable/factor affects:
🔹 Temperature (temp_max, temp_min) → Affects shirt choice and bottom wear
- Hot day? → Short sleeves & shorts
- Chilly morning? → Long sleeves & pants
🌧 Precipitation (precipitation) → Tells us if a raincoat or snow jacket is needed
- Light rain? → Raincoat, but no need for extra layers
- Heavy snow? → Snow jacket & extra warmth
💨 Wind Speed (wind) → Helps decide if a windbreaker is necessary
- Breezy but warm? → No jacket needed
- Strong winds? → Windbreaker, even if it's sunny
☀️ Weather Type (weather) → The AI’s main prediction
- This is the final decision that determines all clothing recommendations.
Instead of making generic weather predictions, the AI looks at multiple factors to give recommendations that actually make sense.
For instance:
- Cold + Rainy + Windy? → The AI suggests a long-sleeve shirt, pants, and a raincoat or windbreaker.
- Warm + Sunny? → The AI suggests a short-sleeve shirt and shorts—no jacket needed.
- Cold + Snowing? → The AI suggests a snow jacket, long-sleeve shirt, and pants.
This means no more standing in front of your closet, wondering if you’ll freeze or sweat through your clothes. The AI does the thinking for you.
Now that we’ve trained our model, it’s time to answer the big question: How well does it actually work?
Our AI’s goal wasn’t just to predict the weather—it also needed to give practical outfit recommendations so people could dress appropriately. Let’s break down how well it performed, what worked, what didn’t, and what could be improved.
We evaluated the model using accuracy, precision, recall, and F1-score, and here’s what we found:
- Strong overall accuracy – The model successfully predicts common weather conditions (Rain, Sun) and significantly improved its ability to detect rare ones (Drizzle, Fog, Snow).
- Balanced precision & recall – After manually oversampling rare weather types, the model was much better at detecting them while still maintaining strong performance for common ones.
- Practical clothing recommendations – The AI reliably suggested jackets on rainy days, shorts on warm days, and windbreakers when it’s windy.
| Weather Condition | Precision | Recall | F1-Score |
|---|---|---|---|
| Rain | 0.98 | 0.88 | 0.93 |
| Sun | 0.86 | 0.88 | 0.87 |
| Drizzle | 0.82 | 0.91 | 0.86 |
| Fog | 0.87 | 0.96 | 0.91 |
| Snow | 0.89 | 0.89 | 0.89 |
- Rain & Sun were predicted very accurately. These are the most common weather types, so the model had plenty of data to learn from.
- Rare weather types (Drizzle, Fog, Snow) improved significantly after balancing the dataset. Originally, the model struggled with these, but now they are detected far more reliably.
- Realistic clothing recommendations. Instead of just predicting the weather, the AI connected the forecast to actual human decisions—choosing short vs. long sleeves, jackets vs. no jackets, and shorts vs. pants.
- Handling Edge Cases (e.g., Light Rain vs. Drizzle) – Some borderline cases were tricky, especially between Drizzle and Rain or Fog and Cloudy conditions. The AI sometimes misclassified these because there isn’t always a clear distinction in the dataset.
- Wind-Based Recommendations – The AI could detect when it was windy, but deciding when wind speed actually makes someone feel cold was a bit more complicated. Right now, the windbreaker recommendation is based only on speed, but humidity and temperature might play a role too.
- Better Seasonal Adjustments – The model doesn't currently factor in seasonality (e.g., Seattle winters are colder than summer). This means that some temperature-based clothing recommendations could be improved by understanding seasonal trends.
- Improve Drizzle & Fog Predictions with More Context
- The AI could use humidity levels or visibility data to better distinguish Fog from Cloudy days.
- Adding rain intensity measurements could help separate Drizzle from Rain more accurately.
- Tune Windbreaker Recommendations
- Instead of recommending a windbreaker based only on wind speed, the model could factor in feels-like temperature (wind chill).
- We could also look at how wind speed affects temperature perception—a 10°C day with strong wind feels much colder than one without wind.
- Introduce Seasonal Awareness
- Right now, the AI treats a 10°C day the same way in summer and winter, but in reality, people dress differently in different seasons.
- Adding month-based adjustments (e.g., people wear jackets earlier in fall than in spring) could improve clothing recommendations.
- Try a More Advanced Model
- While Random Forest worked well, we could try Gradient Boosting (XGBoost) to see if a more refined model improves predictions.
- This could be useful for subtle differences between weather types that Random Forest sometimes struggles with.
- Yes! This AI is already a big step up from just checking the weather—it actually helps people decide what to wear based on the forecast.
- It gets common weather types right, and now it also performs well on rarer conditions like drizzle, fog, and snow.
- There’s still room for improvement, especially in edge cases, wind handling, and seasonal adjustments, but overall, the AI does exactly what it was designed to do—help people dress smarter for the weather.
- With some additional refinements, this AI could be integrated into real-world weather apps to make dressing for the day effortless!
Throughout this project, various tools, datasets, and AI assistance were used to develop, refine, and improve the weather prediction and outfit recommendation model. Below is a detailed list of resources used, ensuring full transparency and proper attribution.
- Seattle Weather Dataset (2012 - 2015), source: https://www.kaggle.com/datasets/ananthr1/weather-prediction/data
- This dataset provided historical weather data, including precipitation, temperature, wind speed, and overall weather conditions
Several open-source Python libraries were used to preprocess the data, build the machine learning model, and generate visualizations:
- Data Processing & Analysis:
pandas— Data handling and manipulationnumpy— Numerical computations
- Machine Learning & Model Training:
scikit-learn— Used for Random Forest classification, model evaluation, and train-test splitting
- Data Visualization
matplotlib— Used for generating plots and heatmapsseaborn— Enhanced data visualization for feature relationships
- Class Imbalance Handling:
- sklearn.utils.resample — Used for manual oversampling of underrepresented weather classes
Role: ChatGPT was used throughout the project to:
- optimize model choices and parameter tuning
- assist in debugging and evaluating model performance
- Standard machine learning documentation and resources were consulted, including:
- Scikit-learn Documentation, source: https://scikit-learn.org/0.21/documentation.html
- Matplotlib Documentation, source: https://matplotlib.org/stable/index.html