Findings and Final Product

Overview

In this section, we summarized the findings of the model training as well as the visitor traffic forecasting. Moreover, we give an overview over the final product built.

The analysis of visitor traffic in the Bavarian Forest National Park utilized the Extra Trees Regressor, focusing on predictions across six distinct regions. We examined 15 target variables, including total visitor counts and IN/OUT values for each region, as well as overall metrics for the park.

---

Model Training

Model Fit Indices

The performance of the Extra Trees Regressor was evaluated using 5-fold cross-validation for both training and test data. The average results across all folds for the target variables are summarized in the table below:

Target/Model	MAE	MSE	RMSE	R²	RMSLE	MAPE
Traffic (IN + OUT for entire Park)	46.1162	8917.1099	92.2189	0.8081	0.8319	1.0861
IN (entire park)	24.3335	2442.0343	48.2164	0.7974	0.7300	1.0028
OUT (entire park)	24.1747	2458.5690	48.4344	0.7931	0.8057	1.0266
Lusen-Mauth-Finsterau (IN)	6.7317	201.3070	14.0569	0.5393	0.8434	1.4217
Lusen-Mauth-Finsterau (OUT)	8.0701	257.9571	15.9973	0.5997	0.9015	1.6323
Nationalparkzentrum Lusen (IN)	7.9054	287.9660	16.6835	0.6963	0.6837	1.1265
Nationalparkzentrum Lusen (OUT)	7.5386	260.5668	15.9148	0.7029	0.7600	1.1740
Rachel-Spiegelau (IN)	5.9471	147.6946	11.5325	0.6194	0.6604	1.0976
Rachel-Spiegelau (OUT)	5.1613	114.6056	10.1086	0.6211	0.6459	1.1172
Falkenstein-Schwellhäusl (IN)	6.3214	175.2889	12.7272	0.7207	0.6538	1.0518
Falkenstein-Schwellhäusl (OUT)	5.9329	159.9409	12.2524	0.7392	0.6455	1.0008
Scheuereck-Schachten-Trinkwassertalsperre (IN)	2.4397	57.1253	6.9169	0.1636	0.6783	1.0380
Scheuereck-Schachten-Trinkwassertalsperre (OUT)	2.1488	21.6679	4.2865	0.3744	0.6556	1.0186
Nationalparkzentrum Falkenstein (IN)	5.0964	116.4658	10.7654	0.6457	0.6387	1.0955
Nationalparkzentrum Falkenstein (OUT)	4.7758	106.0070	10.2023	0.6063	0.6482	1.2331

Understanding the Metrics

Before we analyze the specific results, let's clarify the metrics used:

• MAE (Mean Absolute Error): Measures the average absolute difference between predicted and actual values.
• MSE (Mean Squared Error): Measures the average squared difference between predicted and actual values.
• RMSE (Root Mean Squared Error): The square root of MSE, often preferred for interpretability as it's in the same units as the target variable.
• R² (R-squared): Indicates the proportion of variance in the target variable explained by the model. Higher values are better.
• RMSLE (Root Mean Squared Logarithmic Error): Useful for predicting values on a logarithmic scale (e.g., when values can range from 0 to infinity).
• MAPE (Mean Absolute Percentage Error): Measures the average percentage error between predicted and actual values.

Interpreting the Results

Based on the provided table, here are some observations:

1. Overall Visitor Traffic:

   • The model predicted an average total traffic count (IN + OUT) of 46.12 with a R² value of 0.8081, indicating a strong relationship between predicted and actual values. This suggests that the model successfully captures the dynamics of visitor traffic across the entire park.

2. Regional Performance:

   • Lusen-Mauth-Finsterau showed relatively lower prediction errors, with MAE values of 6.73 (IN) and 8.07 (OUT). These figures suggest a stable visitor count in this region, potentially influenced by its accessibility and visitor amenities.
   • Nationalparkzentrum Lusen also performed well, with MAE values of 7.91 (IN) and 7.54 (OUT), indicating a robust model fit for this popular tourist area.
   • In contrast, the Scheuereck-Schachten-Trinkwassertalsperre region exhibited the lowest prediction accuracy, with MAE values of 2.44 (IN) and 2.15 (OUT). The low R² value of 0.1636 for IN indicates that this region's visitor traffic may be influenced by more unpredictable factors, warranting further investigation.

3. Traffic Trends:

   • The overall IN and OUT values for the park were relatively balanced, with minor variations indicating consistent visitor patterns throughout the year. For example, the predicted IN count of 24.33 for the entire park suggests that visitors are more inclined to enter the park compared to exiting, reflecting a positive visitor experience and prolonged stays.

4. Regional Disparities:

   • Some regions, such as Rachel-Spiegelau and Falkenstein-Schwellhäusl, demonstrated relatively close MAE values (around 5-7 for both IN and OUT), suggesting similar patterns of visitor behavior. The consistent metrics across these regions may indicate comparable attractions or services that draw visitors in and keep them within these areas.

Conclusion

The findings illustrate the Extra Trees Regressor's effectiveness in predicting visitor traffic across various regions of the Bavarian Forest National Park. The results indicate strong predictive power, especially for areas with consistent visitor patterns. These insights can aid park management in resource allocation, event planning, and enhancing visitor experience based on anticipated traffic flows.

---

Visitor Traffic Forecasting for Bavarian Forest National Park (BFNP)

We evaluated the ExtraTree Regressor model on unseen data from May 1, 2024, to July 22, 2024. The model was trained on historical data and used to forecast visitor traffic in the Bavarian Forest National Park (BFNP). The results across various regions and target variables are presented below.

Evaluation on Unseen Data

The graphs presented above show how the model performs on unseen data from May to July 2024, which was not included in the model's training or validation phases. Evaluating on unseen data allows us to assess the model's generalizability and real-world applicability for forecasting future visitor counts.

Comparison of predicted and actual values on unseen data (1)

Comparison of predicted and actual values on unseen data (2)

---

Traffic Abs

The total park traffic (traffic_abs) is consistently overestimated in both low and high-traffic periods. The predicted values follow the general trend of the actual counts but are consistently higher, particularly during low-traffic periods.

Sum IN and Sum OUT

For sum_IN_abs and sum_OUT_abs, the ExtraTree Regressor successfully captures the cyclic nature of visitor inflow and outflow. However, the magnitude of the predicted values often falls short of actual observations during peak traffic days, indicating that the model could struggle to forecast sharp increases in visitor numbers.

Region-Specific Analysis

• Rachel-Spiegelau IN/OUT:

• The model captures the general trend for both inbound and outbound traffic at Rachel-Spiegelau. However, it underestimates the peaks, especially on days of high visitor traffic.

• Scheuereck-Schachten-Trinkwassertalsperre IN/OUT:

• Predictions for IN and OUT traffic at this region show better alignment with actual counts during low-traffic periods. The model slightly underestimates the peaks during high-traffic days, though it captures the overall flow well.

• Falkenstein-Schwellhäusl IN/OUT:

• For both IN and OUT traffic, the model demonstrates its strength in recognizing the cyclic nature of the data. However, peak values are significantly underestimated during high-traffic days, indicating a need for additional model tuning.

• Nationalparkzentrum Lusen IN/OUT:

• The model performs similarly in this region, capturing trends but struggling to forecast peak visitor numbers. There is some overestimation during low-traffic days, but the overall cyclic pattern is well-represented.

Model Performance Insights

General Performance

The ExtraTree Regressor does well in capturing general trends and seasonality across all regions. However, the model struggles with accurately predicting the magnitude of peaks and troughs, especially during periods of high traffic. This is a common issue seen in all regions, suggesting the need for advanced feature engineering or hybrid models that can better capture these extremes.

Predictions for Unseen Data

Evaluating the model on unseen data has revealed the following key insights:

• Cyclic Trends: The model successfully captures recurring patterns in visitor traffic, likely influenced by weekends, holidays, and seasonal effects.

• Underestimation of Peaks: Across almost all regions, the model consistently underestimates peak visitor counts, especially during high-traffic periods.

• Overestimation of Low-Traffic Days: During low-traffic days, the model tends to overestimate the visitor count slightly, leading to higher errors in some regions.

Key Takeaways and Future Work

Improvements to Handle Extremes

The model's underestimation of peak traffic and overestimation during low-traffic periods suggest that further feature engineering or model enhancements could help. Techniques such as gradient boosting or hybrid models that combine tree-based methods with sequence models like LSTMs may be better suited to handling sharp traffic spikes.

Region-Specific Analysis

Different regions exhibit varied model performance, with some regions showing better alignment between actual and predicted values. This variability indicates that additional region-specific features (e.g., local weather patterns, events) may improve the model’s accuracy in certain areas.

Feature Engineering and Hybrid Approaches

Future iterations of the model could benefit from advanced feature engineering. For example, incorporating special event indicators, weather anomalies, or holiday-specific flags could help the model better predict extreme values. Additionally, exploring hybrid approaches (e.g., combining ExtraTree with LSTM models) might improve the overall performance and accuracy.

Feature Importance for Every Target Variable

Feature Importance among every Target Variable

Key Findings:

• Temporal Factors:

  • Hour: The time of day is the most influential factor, suggesting that traffic patterns exhibit strong diurnal variations.

  • Hour_cos and Hour_sin: These features capture cyclical patterns, likely indicating seasonal or weekly trends in visitor traffic.

• Weather Conditions:

  • Temperature (°C): Warmer temperatures correlate with higher visitor counts, suggesting that favorable weather conditions attract more people.
  • Relative Humidity (%): Humidity might also play a role, potentially influencing visitor comfort levels.

• Special Events:

  • Feiertag Bayern: Holidays in Bavaria significantly impact traffic, indicating that special events or celebrations draw more visitors.
  • Schulferien Bayern: School holidays in Bavaria also influence traffic, likely due to increased leisure time for families.

• Weather-Related Factors:

  • ZScore Daily Max Wind Speed (km/h): Wind speed might affect visitor comfort and outdoor activities, influencing traffic patterns.

Interpretation:

• Peak Times: The importance of the "Hour" feature suggests that there are specific peak times during the day when visitor traffic is significantly higher.
• Seasonal Variations: The cyclical patterns captured by "Hour_cos" and "Hour_sin" might indicate seasonal variations in traffic, with higher numbers of visitors during certain months or periods of the year.
• Weather Influence: Favorable weather conditions (e.g., warm and sunny days) might attract more visitors, as evidenced by the importance of temperature and relative humidity.
• Special Event Impact: Holidays and school holidays seem to be significant factors, suggesting that special events or time off from school can drive increased traffic.

---

Final Product

Dashboard Bavarian Forest

Visitor Dashboard

The visitor dashboard provides an intuitive interface that enables users to access essential information in real-time. Visitors can effortlessly view current weather conditions, allowing them to prepare for their day effectively. Additionally, the dashboard showcases hourly predictions for the upcoming week, helping users plan their visits with confidence. For those interested in maximizing their experience, links to various recreational activities are readily available, guiding visitors toward enjoyable options within the park. Moreover, real-time parking occupancy data is displayed, ensuring that users are informed about available parking spaces before arriving. This comprehensive flow of information empowers visitors to make informed decisions, enhancing their overall experience.

Visitor Dashboard

---

Admin Dashboard

The admin dashboard serves as a powerful management tool, enabling administrators to monitor and analyze key metrics related to park operations effectively. Through this interface, admins can access forecasted data on visitor occupancy, providing insights into current crowd levels across the park. Additionally, the dashboard offers detailed visitor predictions for the upcoming week, presented in absolute numbers. This forecasting allows admins to anticipate visitor flow and make informed decisions, ensuring adequate staffing, resource allocation, and visitor experience management. For parking, the dashboard provides real-time occupancy data, displayed as absolute values instead of general categories like low, medium, or high. This level of detail helps administrators optimize parking management and plan ahead for potential surges, ultimately improving the park's operational efficiency and enhancing the visitor experience.

Admin Dashboard

---

Data Accessibility Point

Data Accessibilty point has two main sections that are as follow:

• Upload/Download Functionality: The admin dashboard provides seamless access to data upload and download capabilities. Administrators can easily upload new datasets or update existing ones directly through the interface, ensuring that the system is always using the most up-to-date information. The download functionality allows users to export data related to visitors, parking occupancy, and weather for further analysis or record-keeping. This feature ensures smooth data integration and easy access for offline analysis or reporting.

Dashboard: Data Access Point for Admins

• Querying the Data: Admins can efficiently query the data using advanced filtering options within the dashboard. They can apply filters such as date ranges, specific months, seasons, or years to tailor their data retrieval. Additionally, specific categories like weather, visitor occupancy, and parking can be queried to get detailed insights based on the selected criteria. This flexible querying system allows administrators to gather precise, actionable data for analysis and operational planning.

Dashboard: Data Querying Section in the Data Access Point

---

Conclusion

The results of our predictive models demonstrated strong performance on unseen data, with MAPE(mean absolute percentage error) indicating reliable accuracy across the various target variables. The feature importance analysis provided valuable insights, highlighting the key factors driving visitor flows and parking occupancy, which are crucial for making data-driven decisions.

The culmination of these efforts is the deployment of an interactive dashboard that not only leverages predictive analytics but also integrates real-time data to offer actionable insights. The combination of accurate visitor predictions, real-time parking availability, and weather forecasting enhances both the visitor experience and the administrative decision-making process. Overall, the solution provides a powerful tool for optimizing park management and ensuring smooth operations, making the park more accessible and enjoyable for visitors.