Milestone 2

Traffic Pattern Analysis: Milestone 2 Report

In Milestone 2, we accomplished the following objectives:

• Select ML Model: Successfully implemented the Dynamic Time Warping (DTW) model, allowing us to compare time series data effectively, even when the sequences have different lengths or time warps.

• Data Analysis: Conducted comprehensive data analysis, filtering out pedestrian and bicycle entries to focus solely on car and truck tracks. This process ensured a clean representation of vehicular traffic patterns.

• Plotting and Visualization: Utilized tools like Tableau for in-depth data visualization, aiding us in understanding complex patterns and trends.

Dynamic Time Warping (DTW) Model: DTW is a robust distance measure enabling the comparison of time series data, accommodating variations in length and time warping. By incorporating features like heading, xVelocity, yVelocity, xAcceleration, xCenter, and yCenter, we classify clusters based on direction, enriching our understanding of traffic flow patterns.

Anomaly Detection: In progress, exploring various methods to detect anomalies. Currently, we identify zero-speed anomalies and U-turn anomalies using specific criteria at different intersections and tracks.

• Cleaned Data: Filtered out pedestrian and bicycle entries, focusing solely on car and truck tracks, ensuring a more accurate representation of vehicular traffic.

Classified Data Based on Days:

• Velocity Trends: Mondays exhibit the highest average velocity, suggesting smoother traffic flow. Conversely, Wednesdays showcase the slowest traffic, possibly due to congestion. Tuesdays and Thursdays fall in between, indicating moderate traffic speeds.

• Acceleration Trends: Mondays feature the highest average acceleration, potentially indicating more aggressive driving. Wednesdays display stable acceleration, while Tuesdays and Thursdays show balanced driving behavior.

Milestone 2 Feedback:

• Lessons Learned: Recognized the challenges in finding anomalies within relatively clean data, prompting the exploration of alternative anomaly detection logics.

• Challenges: Encountered difficulties in training ML models to find outliers, emphasizing the importance of selecting appropriate features, and experimenting with different anomaly detection methods.