• Goal: Improve visitor satisfaction, resource management, and safety.
• Feedback Adjustments:
  • Focused on classification modeling (isOpenToPublic).
• Refined Plan:
  • Data Preprocessing: Handle missing values, encode categorical variables.
  • Model: Deep Learning (DNN).
  • Evaluation: AUC, Confusion Matrix.
  • Feature Importance: H2O’s SHAP analysis.
• Peer Feedback:
  • Feedback: Use confusion matrices for error analysis.
  • Response: Analyzed false positives and negatives using confusion matrices.

1. Data Preprocessing: -Missing values were handled via mean imputation for numerical variables and “Unknown” for categorical variables. -Categorical variables were label-encoded. -Logical variables were converted to binary numeric representation.

2. Modeling Approach: -Focused on building a Deep Learning Model (DNN) using H2O. -Emphasis on ensuring a robust training-validation-testing split.

3. Evaluation Metrics: -Area Under Curve (AUC) for classification performance. -Confusion Matrix to identify classification errors (false positives and false negatives).

4. Feature Importance: -Utilized built-in H2O functionality to rank predictors by importance. -Visualized feature significance to identify key drivers for the target variable.

• Dataset Description:
  • Source: National Park Services API.
  • Observations: 29,754 rows.
  • Features: 45.
  • Target Variable: isOpenToPublic (binary: Open/Closed).

1) Data Cleaning and Feature Engineering: Missing values in numerical columns were replaced with the mean. Missing values in categorical columns were replaced with “Unknown”. Logical columns were replaced with FALSE for missing values. Unnecessary columns were dropped to reduce noise.

2) Model Training: Model that was used for training using H2O: Deep Learning Model (DNN) Data was split into training, validation, and testing sets.

3) Evaluation: Model performance metrics were calculated on the validation and testing datasets. SHAP plot was used for interpretability.

Performance Metrics:

MSE: 0.0611

RMSE: 0.2472

AUC: 0.5239

LogLoss: 0.2407

Interpretability: SHAP values indicate Top Predictors.

1. Top Features:

   • park_states: The most influential feature in the model. It has a significant positive impact on the model’s prediction.

   • isInteragencyPassAccepted: This feature has a negative contribution when its value is 0.

2. Minor Features:
   • Features like campsites.tentOnly, park_name, and visitor_center_isPassportStampLocation contribute less significantly to the model output compared to others.
3. Actionable Insights:
   • Focus on the most influential features (park_states and activity_name) to better understand the model’s decision-making process.

Findings:

-Parks with a wider range of amenities are more frequently open to the public.Basic amenities such as toilets and potable water have the strongest association with public openness, as they directly affect visitor convenience.

-Parks managed directly by the National Park Service (park_isManagedByNps = 1) are more consistently open to the public compared to those managed by other entities.

-Fee-free parks (isFeeFreePark = TRUE) is more likely to be open to the public.