ISSS608 2017-18 T3 Assign Li Hongxin Methodology

VAST Mini Challenge 1: "Cheep" Shots?

Background

Methodology

Answers

Conclusions

Tools

a. R: used for data cleaning.

Packages: tidyverse

b. Tableau: used for Map & Pattern visualization.

c. Python: used for density visualization, audio visualization and audio classification.

Packages: os, glob, pandas, numpy, matplotlib, seaborn, librosa, sklearn

Process for Data Preparation

The following are 5 key steps for data cleaning, and data manipulation for further visualization and analysis.

• Step 1: Deal with Missing Values. Replace all symbols such as "?", "??:??" in Time, and "No score" in Quality which stand for missing values, into NA.

• Step 2: Fix Data Quality Issues. Transform all letters into uppercase for convenience, and remove extra spaces and "?".

• Step 3: Unify the Date & Time Format. Transform all Date into "%Y-%m-%d" format. If the raw data doesn't contain month or day info, we impute the data as "-01-"(January) and "-01"(the first day). Transform all Time into "HH:mm" format and use standardized all times into 24 hour formatting. If raw data doesn't contain minute info, set it as "00". If raw data contain letters such as "morning", or "dawning", imputed them into "08:00" or "18:00".

• Step 4: Modify Data Types. Change X and Y coordinate from character into int.

• Step 5: Create Season and Timeslot variables based on Date and Time. For example, set March to May as Spring ,and set 06:00 to 12:00 as "Morning".

Pattern Visualization and Analysis

Approach	Description
Geo-spatial Visualization Figure out the effect of Dumping site by focusing on the geographical distribution of birds	i. Scatter Plot on Map Scatter plot contains XY coordinates, which indicates the geographical distribution of birds Create a 200x200 map background to integrate the real geo situation with wildlife reserve Highlight of the Dumping site to help analyse the effect on the birds. Re-scale the XY coordinate from 1x1 gird into 5x5 to show density ii. Kernel Density Plot Apply kernel density plot by year(from 2012 to 2017) on Rose-crested Blue Pipit Add map ground and highlight of Dumping site Create a gif file to show the dynamic change of dense clusters
Trend Visualization Understand the growth pattern of birds	Area/Line Graph Remove the data in 2018. Since only 3-month data available and it was not comparable with other years with 12-month data Create area/line graph of all species and by bird species from 1983 to 2017
Interactive Dashboard Understand the pattern of all 19 bird species, not only the Rose-crested Blue Pipit	Combine the result of Geo-spatial visualization and trend visualization Dashboard includes scatter plot with map background and area graph Users selection: Year/Season/Timeslot(eg.Morning)/Bird Species

Audio Data Analysis

Approach	Description
Audio Visualization	i. Waveplot Plot one "Call" sample and one "Song" sample for each of 19 class since different types of voice may have various patterns as train set Only samples with "Quality A" were selected to avoid noise disturbance Plot the 15 test birds as well and compare with the train set ii. Specgram Plot specgram for the same train set and test birds, to help improve the visualization of bird songs
Audio Classification	i. Feature Extraction 5 types of features were selected, and by combining them we got total 193 features for each bird song Mel-frequency cepstral coefficients (MFCC) Chromagram of a short-time Fourier transform Mel-scaled power spectrogram Octave-based spectral contrast Tonnetz ii. Data Partition and Feature Labels Out of 2081 bird songs, set 70% as training data and 30% as test data Add labels which stand for the name of bird species to each feature iii. Classification Methods Logistic Regression SVM Random Forest