ISSS608 2017-18 T3 Assign Yejerla Amrutha Rajeshwari dp

VAST Mini Challenge 1: "Cheep" Shots?

Context

Data Preparation & Methodology

Data Exploration: Formulating Hypothesis

Bird's-eye View

What's in a Call?

Conclusion

Methodology

Data is augmented with additional columns to facilitate drill down and pattern identification. Following two step methodology is followed to solve the problem statement:

1. Tableau Interactive Dashboard is created as a tool to exercise scepticism about data and helps in formulating hypothesis and sets direction for research to be performed.
2. Resultant hypothesis are pursued using visualisation and statistical techniques. Based on the evidence that these techniques produce, hypothesis are either approved are rejected.

Tools Used

Tableau: Exploratory Data Analysis
R packages: Custom Visualizations and Audio Processing

1. Audio Processing: warbleR, bioacoustics
2. Data handling: lubridate, chron, tidyr, dplyr, stringr
3. Visualization: gplots, ggplot2, RColorBrewer

Data Pre-Processing

Pre-processing Data table: Metadata provided for the recordings consists of not applicable values and estimates. Following data preparation steps are performed in R.

• Correcting date format discrepancies.
• Setting non-repetitive factor levels for Vocalization_type column.
• Formatting Time column so as to follow %H:%M:%S format. Few of the time records follow 24-hr format and few others have AM and PM.
• Creating new features:
  • Week, Month, Year columns are created. Though these columns do not have direct relevance to the behaviour of birds, these extra columns facilitate drilldown.
  • Birds behaviour varies with seasons. Following American seasons
    • Spring: March 1 to May 31
    • Fall: June 1 to August 31
    • Summer: September 1 to November 30
    • Winter: December 1 to February 28
  • modified_year: This column is created to facilitate creation of graphs. For any year, winter spans January, February and also December. When plotting graphs for each season-year using the year column, records for all these months are aggregated together and is not desirable. To mitigate this, December of an year is considered as a part of winter of the next year to have continuity in data representation.

Pre-processing Audio files: Pre-processing is done on the classified bird recordings as well as test recordings from Kaisos. Following are the pre-processing steps followed:

• Conversion of files from mp3 format to wav format to facilitate processing in R.
• Recordings typically have background noise and capture calls and songs from unintended birds. Signal detection is performed on the recordings using autodetec function available in wableR package and individual components are derived.
• Only few components are retained based on the Signal to Noise Ratio. Further insights will be derived based on these components.