DATA ANALYST

CLIMATE WINS

Tools

 

  • Microsoft Excel

  • Microsoft Word

  • Microsoft PowerPoint

  • Python

  • Tableau

Skills

 

  • Data Extraction and Cleaning.

  • Data Exploration.

  • Data Preprocessing.

  • Machine Learning with Python

  • Joining tables

Data

 

  • Dataset Weather Prediction

  • Project brief

This is a Predicting Weather Variations with Machine Learning for ClimateWins a European nonprofit organization. Cimate Wins is interested in using machine learning to help predict the consequences of climate change around Europe and, potentially, the world.

 

Objective

 

Identify weather patterns outside the regional norm in Europe. Determine if unusual weather patterns are increasing. Generate possibilities for future weather conditions over the next 25 to 50 years. Determine the safest regions for habitation in Europe over the next 25 to 50 years.

Hypothesis: CNNs can better interpret radar and satellite imagery to classify weather conditions, improving the prediction of weather trends.

 

Approach:

• Developed a CNN model to classify radar images of various

weather conditions (e.g., cloudy, rainy, sunny).

• Used Bayesian optimization to refine hyperparameters like the

number of neurons, batch size, and learning rate for better

accuracy.

 

Model Used: CNN with Bayesian optimization.

 

Result:

• Initial Accuracy: The unoptimized CNN achieved around 11% accuracy.

 

• Optimized Model Accuracy: After Bayesian optimization, accuracy improved significantly to 80%.

 

• Confusion Matrix: Showcases the model’s ability to differentiate between weather conditions such as 'cloudy' and 'rainy,' highlighting areas where classification errors reduced after optimization.

 

Conclusion: This model demonstrated the potential for analyzing complex visual data, making it useful for predicting shifts in weather patterns.

Hypothesis: A random forest model can identify abnormal

weather trends based on historical patterns and station data.

 

Approach:

• Used RandomForestClassifier to analyze key features like

precipitation and temperature across various European

weather stations.

• Applied RandomizedSearchCV  for hyperparameter tuning,

optimizing parameters such as  n_estimators, max_depth, and

min_samples_split to  improve model accuracy.

 

Model Used: RandomForestClassifier  with optimized hyperparameters.

 

Result:

• Accuracy: Improved from an initial 71.2% to approximately

72% after hyperparameter optimization.

 

• Feature Importance: The most predictive features were

Kassel, Belgrade and Heathrow stations, highlighting the areas with significant weather variation.

 

Conclusion: This approach successfully identified areas

experiencing deviations from historical patterns, helping to

detect increasing anomalies like shifts in precipitation and

temperature extremes.

Lost Plot

Displays training and

validation loss. Lower loss

indicates better fit, with spikes showing areas for

improvement.

Tracks accuracy over epochs. Higher values indicate better performance; fluctuations suggest varying model stability.

Shows how well the model

classifies weather types. Most predictions are accurate, but some classes are confused with others.

Accuracy Plot

Confusion Matrix

Model Performance Evaluation Using CNN

Deep Learning for Image-Based Weather Classification

Deep Learning with CNNs

 

• Description: CNNs were applied to classify weather conditions based on radar and satellite imagery.

 

• Results: Improved test accuracy to 80.17% through Bayesian optimization.

 

• Key Features: Enabled analysis of complex spatial patterns in weather images.

Random Forest Model

 

• Description: An ensemble learning method used to classify weather conditions, predict safe flight conditions, and analyze variable importance.

 

• Results: Achieved 73% accuracy with Randomized Search CV for multi-station data; 100% accuracy when focused on a single station like Maastricht.

 

• Key Features: Precipitation, temperature metrics,

cloud cover, and sunshine.

Predicting Anomalies with Random Forest

Predicting Weather Anomalies with Random Forest: Focus on using ensemble models to analyze shifts in weather patterns based on historical data.

Use Random Forest models for immediate analysis of feature importance and to identify key predictors of abnormal weather patterns.

 

Implement CNNs for analyzing satellite data and weather imagery to improve real-time classification of weather conditions.

 

Invest in developing GANs for longer-term scenario simulation, helping ClimateWins plan for potential future climates.

RECOMMENDATIONS

Ivonne Aspilcueta 

Data Analyst

 

Hermosa Beach, CA, United States