Global Landslide Catalog Analysis
Shailja Kartik
December 06, 2021
Introduction
The dataset we are using for my Final Project is the “Global Landslide Catalog Export” dataset obtained from NASA’s Open Data Portal.
A summary of the dataset is available at: https://data.nasa.gov/Earth-Science/Global-Landslide-Catalog-Export/dd9e-wu2v
The downloadable version of the dataset in the CSV format is from: https://catalog.data.gov/dataset/global-landslide-catalog-export
To learn more about NASA’s work on the landslide, please visit the homepage: https://gpm.nasa.gov/landslides/index.html
Motivation behind using the Global Landslide Catalog Export (GLC)
Landslides cause billions of dollars in infrastructural damage and thousands of deaths worldwide. Data on past landslide events guides future disaster prevention, but we do not have a global picture of exactly when and where landslides occur.
NASA scientists have been building an open global inventory of landslides to address this problem. Knowing where and when landslides occur can help communities worldwide prepare for these disasters.
Below figure is just an overview of the collected landslides so far:
Through this project, we would welcome an opportunity to help make an informed decision that could save lives and property
About the dataset
The Global Landslide Catalog (GLC) has been compiled since 2007 at NASA Goddard Space Flight Center.
The GLC is a collection of observational studies.
The GLC was developed to identify rainfall-triggered landslide events worldwide, regardless of size, impact, or location.
The GLC considers all types of mass movements triggered by rainfall, which have been reported in the media, disaster databases, scientific reports, or other sources.
The dataset contains 31 Attributes and has 11033 observations.
Each observation is a Landslide.
The list of all the Variables which are available for us to explore along with their types are listed below:
knitr::kable(attributes, "simple", col.names = c("Attribute Names", "Attribute Type"), align = c("l", "c"), caption = "Variables of the Dataset and their Types")| Attribute Names | Attribute Type |
|---|---|
| source_name | Text |
| source_link | Website URL |
| event_id | Number |
| event_date | Date & Time |
| event_time | Date & Time |
| event_title | Text |
| event_description | Text |
| location_description | Text |
| location_accuracy | Text |
| landslide_category | Text |
| landslide_trigger | Text |
| landslide_size | Text |
| landslide_setting | Text |
| fatality_count | Number |
| injury_count | Number |
| storm_name | Text |
| photo_link | Website URL |
| notes | Text |
| event_import_source | Text |
| event_import_id | Number |
| country_name | Text |
| country_code | Text |
| admin_division_name | Text |
| admin_division_population | Number |
| gazeteer_closest_point | Text |
| gazeteer_distance | Number |
| submitted_date | Date & Time |
| created_date | Date & Time |
| last_edited_date | Date & Time |
| longitude | Number |
| latitude | Number |
However,we will be focusing only on the below Variables for my Project Analysis:
knitr::kable(attributes_interest, "simple", col.names = c("Variable Names", "Variable Type"), align = c("l", "c"), caption = "Variables of the Dataset we will explore")| Variable Names | Variable Type |
|---|---|
| source_name | Text |
| event_id | Number |
| event_date | Date & Time |
| event_time | Date & Time |
| event_title | Text |
| event_description | Text |
| location_description | Text |
| location_accuracy | Text |
| landslide_category | Text |
| landslide_trigger | Text |
| landslide_size | Text |
| fatality_count | Number |
| injury_count | Number |
| storm_name | Text |
| country_name | Text |
| country_code | Text |
| admin_division_name | Text |
| admin_division_population | Number |
| longitude | Number |
| latitude | Number |
Goal
Through the below data analysis, we want to answer these questions:
- How many people were killed in the largest landslide ever recorded?
- Are the sizes of various landslides equally distributed in the dataset?
- What are the countries with more than 50 injured recorded in any landslide?
- Is there any correlation between the numerical variable?
- Perform hypothesis testing to see if the mean of the fatality_count of any two countries with the same number of landslides will be the same or not
- Use Logistic Regression to predict the size of the landslide
Analysis
Loading the Package and Data
The first step is to load the necessary library, which contains all the datasets
If the package is not installed, please uncomment the below line of R code and execute it on your machine. The below statement installs the pre-requisite package
#install.packages("tidyverse")
#install.packages("ggplot2")
#install.packages("maps")Now that the package is installed, we need to load the required library
# Loading required libraries #
library(tidyverse)## Warning: package 'tidyverse' was built under R version 4.1.2## -- Attaching packages --------------------------------------- tidyverse 1.3.1 --## v ggplot2 3.3.5 v purrr 0.3.4
## v tibble 3.1.4 v dplyr 1.0.7
## v tidyr 1.1.3 v stringr 1.4.0
## v readr 2.0.1 v forcats 0.5.1## Warning: package 'ggplot2' was built under R version 4.1.2## -- Conflicts ------------------------------------------ tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()library(ggplot2)In the next step, we will create a local object that will hold our dataset.
The name of our local object will be “globalLandslide_data”.
globalLandslide_data <- as_tibble(read.csv("D:\\SJSU\\1stSem\\Study\\ISE-201\\Submissions\\ProjectProposal-2\\originaldataset\\Global_Landslide_Catalog_Export.csv"))Examining the Data
This section will use some basic steps to examine our data. Here we will also see if any changes are required in our dataset to make our work easy and smooth.
This stage can also be referred to as “Cleaning the Data.”
1. Evaluating the structure of the data
str(globalLandslide_data)## tibble [11,033 x 31] (S3: tbl_df/tbl/data.frame)
## $ source_name : chr [1:11033] "AGU" "Oregonian" "CBS News" "Reuters" ...
## $ source_link : chr [1:11033] "https://blogs.agu.org/landslideblog/2008/10/14/the-lifan-landslide-from-natural-disaster-to-cover-up/" "http://www.oregonlive.com/news/index.ssf/2009/01/landslide_plows_through_lake_o.html" "https://www.cbsnews.com/news/dozens-missing-after-peru-landslides/" "https://in.reuters.com/article/idINIndia-41450420090731" ...
## $ event_id : int [1:11033] 684 956 973 1067 2603 4203 4290 225 236 873 ...
## $ event_date : chr [1:11033] "08/01/2008 12:00:00 AM" "01/02/2009 02:00:00 AM" "01/19/2007 12:00:00 AM" "07/31/2009 12:00:00 AM" ...
## $ event_time : logi [1:11033] NA NA NA NA NA NA ...
## $ event_title : chr [1:11033] "Sigou Village, Loufan County, Shanxi Province" "Lake Oswego, Oregon" "San Ramon district, 195 miles northeast of the capital, Lima, " "Dailekh district" ...
## $ event_description : chr [1:11033] "occurred early in morning, 11 villagers buried in 7 houses" "Hours of heavy rain are to blame for an overnight mudslide in Lake Oswego. " "(CBS/AP) At least 10 people died and as many as 80 were still missing Wednesday in central Peru after torrentia"| __truncated__ "One person was killed in Dailekh district, police said." ...
## $ location_description : chr [1:11033] "Sigou Village, Loufan County, Shanxi Province" "Lake Oswego, Oregon" "San Ramon district, 195 miles northeast of the capital, Lima, " "Dailekh district" ...
## $ location_accuracy : chr [1:11033] "unknown" "5km" "10km" "unknown" ...
## $ landslide_category : chr [1:11033] "landslide" "mudslide" "landslide" "landslide" ...
## $ landslide_trigger : chr [1:11033] "rain" "downpour" "downpour" "monsoon" ...
## $ landslide_size : chr [1:11033] "large" "small" "large" "medium" ...
## $ landslide_setting : chr [1:11033] "mine" "unknown" "unknown" "unknown" ...
## $ fatality_count : int [1:11033] 11 0 10 1 0 0 0 3 NA 2 ...
## $ injury_count : int [1:11033] NA NA NA NA NA NA NA NA NA NA ...
## $ storm_name : chr [1:11033] "" "" "" "" ...
## $ photo_link : chr [1:11033] "" "" "" "" ...
## $ notes : chr [1:11033] "" "" "" "" ...
## $ event_import_source : chr [1:11033] "glc" "glc" "glc" "glc" ...
## $ event_import_id : num [1:11033] 684 956 973 1067 2603 ...
## $ country_name : chr [1:11033] "China" "United States" "Peru" "Nepal" ...
## $ country_code : chr [1:11033] "CN" "US" "PE" "NP" ...
## $ admin_division_name : chr [1:11033] "Shaanxi" "Oregon" "JunÃn" "Mid Western" ...
## $ admin_division_population: int [1:11033] 0 36619 14708 20908 798634 2404 2126 3191 2689 0 ...
## $ gazeteer_closest_point : chr [1:11033] "Jingyang" "Lake Oswego" "San Ramón" "Dailekh" ...
## $ gazeteer_distance : num [1:11033] 41.021 0.603 0.855 0.754 2.022 ...
## $ submitted_date : chr [1:11033] "04/01/2014 12:00:00 AM" "04/01/2014 12:00:00 AM" "04/01/2014 12:00:00 AM" "04/01/2014 12:00:00 AM" ...
## $ created_date : chr [1:11033] "11/20/2017 03:17:00 PM" "11/20/2017 03:17:00 PM" "11/20/2017 03:17:00 PM" "11/20/2017 03:17:00 PM" ...
## $ last_edited_date : chr [1:11033] "02/15/2018 03:51:00 PM" "02/15/2018 03:51:00 PM" "02/15/2018 03:51:00 PM" "02/15/2018 03:51:00 PM" ...
## $ longitude : num [1:11033] 107.5 -122.7 -75.4 81.7 123.9 ...
## $ latitude : num [1:11033] 32.6 45.4 -11.1 28.8 10.3 ...The output of the above query tells us the structure of the dataset.
We have a data frame with 11,033 observations on 31 variables. And the dataset is a mix of categorical, nominal, numerical, and continuous variables.
2. Peeking at the data
Looking at the first few observations in the dataframe
head(globalLandslide_data)## # A tibble: 6 x 31
## source_name source_link event_id event_date event_time event_title
## <chr> <chr> <int> <chr> <lgl> <chr>
## 1 AGU https://blog~ 684 08/01/200~ NA "Sigou Vill~
## 2 Oregonian http://www.o~ 956 01/02/200~ NA "Lake Osweg~
## 3 CBS News https://www.~ 973 01/19/200~ NA "San Ramon ~
## 4 Reuters https://in.r~ 1067 07/31/200~ NA "Dailekh di~
## 5 The Freeman http://www.p~ 2603 10/16/201~ NA "sitio Baki~
## 6 BusinessWorld Online http://www.b~ 4203 02/16/201~ NA "Paguite, A~
## # ... with 25 more variables: event_description <chr>,
## # location_description <chr>, location_accuracy <chr>,
## # landslide_category <chr>, landslide_trigger <chr>, landslide_size <chr>,
## # landslide_setting <chr>, fatality_count <int>, injury_count <int>,
## # storm_name <chr>, photo_link <chr>, notes <chr>, event_import_source <chr>,
## # event_import_id <dbl>, country_name <chr>, country_code <chr>,
## # admin_division_name <chr>, admin_division_population <int>, ...Looking at the last few observations in the dataset
tail(globalLandslide_data)## # A tibble: 6 x 31
## source_name source_link event_id event_date event_time event_title
## <chr> <chr> <int> <chr> <lgl> <chr>
## 1 St. Maries G~ http://www.gazet~ 10518 03/23/2017~ NA Mudslide abov~
## 2 The Jakarta ~ http://www.theja~ 11109 04/01/2017~ NA Major landsli~
## 3 Greater Kash~ http://www.great~ 10845 03/25/2017~ NA Barnari Sigdi~
## 4 NBC Daily http://www.nbcda~ 10973 12/15/2016~ NA Landslide at ~
## 5 AGU Landslid~ http://blogs.agu~ 10901 04/29/2017~ NA Mayor landsli~
## 6 The Times of~ https://timesofi~ 10949 03/13/2017~ NA Kondapur Comm~
## # ... with 25 more variables: event_description <chr>,
## # location_description <chr>, location_accuracy <chr>,
## # landslide_category <chr>, landslide_trigger <chr>, landslide_size <chr>,
## # landslide_setting <chr>, fatality_count <int>, injury_count <int>,
## # storm_name <chr>, photo_link <chr>, notes <chr>, event_import_source <chr>,
## # event_import_id <dbl>, country_name <chr>, country_code <chr>,
## # admin_division_name <chr>, admin_division_population <int>, ...3. Checking the summary of the Tibble
summary(globalLandslide_data)## source_name source_link event_id event_date
## Length:11033 Length:11033 Min. : 1 Length:11033
## Class :character Class :character 1st Qu.: 2785 Class :character
## Mode :character Mode :character Median : 5563 Mode :character
## Mean : 5599
## 3rd Qu.: 8435
## Max. :11221
##
## event_time event_title event_description location_description
## Mode:logical Length:11033 Length:11033 Length:11033
## NA's:11033 Class :character Class :character Class :character
## Mode :character Mode :character Mode :character
##
##
##
##
## location_accuracy landslide_category landslide_trigger landslide_size
## Length:11033 Length:11033 Length:11033 Length:11033
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
##
## landslide_setting fatality_count injury_count storm_name
## Length:11033 Min. : 0.000 Min. : 0.000 Length:11033
## Class :character 1st Qu.: 0.000 1st Qu.: 0.000 Class :character
## Mode :character Median : 0.000 Median : 0.000 Mode :character
## Mean : 3.219 Mean : 0.752
## 3rd Qu.: 1.000 3rd Qu.: 0.000
## Max. :5000.000 Max. :374.000
## NA's :1385 NA's :5674
## photo_link notes event_import_source event_import_id
## Length:11033 Length:11033 Length:11033 Min. :-111.2
## Class :character Class :character Class :character 1st Qu.:2386.5
## Mode :character Mode :character Mode :character Median :4773.0
## Mean :4798.6
## 3rd Qu.:7189.5
## Max. :9669.0
## NA's :1562
## country_name country_code admin_division_name
## Length:11033 Length:11033 Length:11033
## Class :character Class :character Class :character
## Mode :character Mode :character Mode :character
##
##
##
##
## admin_division_population gazeteer_closest_point gazeteer_distance
## Min. : 0 Length:11033 Min. : 0.000
## 1st Qu.: 1963 Class :character 1st Qu.: 2.364
## Median : 7365 Mode :character Median : 6.255
## Mean : 157760 Mean : 11.874
## 3rd Qu.: 34021 3rd Qu.: 15.816
## Max. :12691836 Max. :215.449
## NA's :1562 NA's :1562
## submitted_date created_date last_edited_date longitude
## Length:11033 Length:11033 Length:11033 Min. :-179.98
## Class :character Class :character Class :character 1st Qu.:-107.87
## Mode :character Mode :character Mode :character Median : 19.69
## Mean : 2.52
## 3rd Qu.: 93.95
## Max. : 179.99
##
## latitude
## Min. :-46.77
## 1st Qu.: 13.92
## Median : 30.53
## Mean : 25.88
## 3rd Qu.: 40.87
## Max. : 72.63
## Quality Check and Data Cleaning
In this section, we will be checking the quality of our data and cleaning our dataset, if required.
1. Checking the Variable Name
# Checking the names of the Columns
names(globalLandslide_data)## [1] "source_name" "source_link"
## [3] "event_id" "event_date"
## [5] "event_time" "event_title"
## [7] "event_description" "location_description"
## [9] "location_accuracy" "landslide_category"
## [11] "landslide_trigger" "landslide_size"
## [13] "landslide_setting" "fatality_count"
## [15] "injury_count" "storm_name"
## [17] "photo_link" "notes"
## [19] "event_import_source" "event_import_id"
## [21] "country_name" "country_code"
## [23] "admin_division_name" "admin_division_population"
## [25] "gazeteer_closest_point" "gazeteer_distance"
## [27] "submitted_date" "created_date"
## [29] "last_edited_date" "longitude"
## [31] "latitude"Our variable names are meaningful, which is great. In addition, the clear variable names help us know the feature we are working on.
2. Removing the Columns which we are not including in our analysis
Let us remove the columns which we are not going to focus on for our further analysis:
globalLandslide_df <- globalLandslide_data[!(colnames(globalLandslide_data) %in% c("source_link", "landslide_setting", "photo_link", "notes",
"event_import_source", "event_import_id", "gazeteer_closest_point", "gazeteer_distance", "submitted_date", "created_date",
"last_edited_date"))]Looking at the structure of the dataframe we will be proceeding with:
str(globalLandslide_df)## tibble [11,033 x 20] (S3: tbl_df/tbl/data.frame)
## $ source_name : chr [1:11033] "AGU" "Oregonian" "CBS News" "Reuters" ...
## $ event_id : int [1:11033] 684 956 973 1067 2603 4203 4290 225 236 873 ...
## $ event_date : chr [1:11033] "08/01/2008 12:00:00 AM" "01/02/2009 02:00:00 AM" "01/19/2007 12:00:00 AM" "07/31/2009 12:00:00 AM" ...
## $ event_time : logi [1:11033] NA NA NA NA NA NA ...
## $ event_title : chr [1:11033] "Sigou Village, Loufan County, Shanxi Province" "Lake Oswego, Oregon" "San Ramon district, 195 miles northeast of the capital, Lima, " "Dailekh district" ...
## $ event_description : chr [1:11033] "occurred early in morning, 11 villagers buried in 7 houses" "Hours of heavy rain are to blame for an overnight mudslide in Lake Oswego. " "(CBS/AP) At least 10 people died and as many as 80 were still missing Wednesday in central Peru after torrentia"| __truncated__ "One person was killed in Dailekh district, police said." ...
## $ location_description : chr [1:11033] "Sigou Village, Loufan County, Shanxi Province" "Lake Oswego, Oregon" "San Ramon district, 195 miles northeast of the capital, Lima, " "Dailekh district" ...
## $ location_accuracy : chr [1:11033] "unknown" "5km" "10km" "unknown" ...
## $ landslide_category : chr [1:11033] "landslide" "mudslide" "landslide" "landslide" ...
## $ landslide_trigger : chr [1:11033] "rain" "downpour" "downpour" "monsoon" ...
## $ landslide_size : chr [1:11033] "large" "small" "large" "medium" ...
## $ fatality_count : int [1:11033] 11 0 10 1 0 0 0 3 NA 2 ...
## $ injury_count : int [1:11033] NA NA NA NA NA NA NA NA NA NA ...
## $ storm_name : chr [1:11033] "" "" "" "" ...
## $ country_name : chr [1:11033] "China" "United States" "Peru" "Nepal" ...
## $ country_code : chr [1:11033] "CN" "US" "PE" "NP" ...
## $ admin_division_name : chr [1:11033] "Shaanxi" "Oregon" "JunÃn" "Mid Western" ...
## $ admin_division_population: int [1:11033] 0 36619 14708 20908 798634 2404 2126 3191 2689 0 ...
## $ longitude : num [1:11033] 107.5 -122.7 -75.4 81.7 123.9 ...
## $ latitude : num [1:11033] 32.6 45.4 -11.1 28.8 10.3 ...We will be proceeding with globalLandslide_df, which has 11,033 observations on 20 variables.
3. Checking for missing data
# Calculate the total numbers of "Not Available" data
sum(is.na(globalLandslide_df))## [1] 19656There are many cells with missing data in our dataset. Let’s see which columns do not have data in them
names(which(colSums(is.na(globalLandslide_df)) > 0))## [1] "event_time" "fatality_count"
## [3] "injury_count" "country_code"
## [5] "admin_division_population"4. Handling Missing Data
Let us address the above columns with missing data one-by-one
# Handling missing data in "injury_count" column
globalLandslide_df$injury_count[is.na(globalLandslide_df$injury_count)] = 0
# Handling missing data in "admin_division_population" column
globalLandslide_df$admin_division_population[is.na(globalLandslide_df$admin_division_population)] = 0
# Handling missing data in "fatality_count" column
globalLandslide_df[!is.na(globalLandslide_df$fatality_count),]## # A tibble: 9,648 x 20
## source_name event_id event_date event_time event_title event_descripti~
## <chr> <int> <chr> <lgl> <chr> <chr>
## 1 AGU 684 08/01/200~ NA "Sigou Vil~ "occurred early~
## 2 Oregonian 956 01/02/200~ NA "Lake Oswe~ "Hours of heavy~
## 3 CBS News 973 01/19/200~ NA "San Ramon~ "(CBS/AP) At le~
## 4 Reuters 1067 07/31/200~ NA "Dailekh d~ "One person was~
## 5 The Freeman 2603 10/16/201~ NA "sitio Bak~ "Another landsl~
## 6 BusinessWorld Online 4203 02/16/201~ NA "Paguite, ~ "Thursday’s l~
## 7 The Spokesman-Review 4290 03/30/201~ NA "Pend Orei~ "In Pend Oreill~
## 8 Crónica Diaria 225 09/02/200~ NA "3 killed ~ "3 killed, incl~
## 9 UPI 873 11/01/200~ NA "Lincang C~ "The report sai~
## 10 BBC News 874 11/01/200~ NA "Kunming, ~ "Yunnan has so ~
## # ... with 9,638 more rows, and 14 more variables: location_description <chr>,
## # location_accuracy <chr>, landslide_category <chr>, landslide_trigger <chr>,
## # landslide_size <chr>, fatality_count <int>, injury_count <dbl>,
## # storm_name <chr>, country_name <chr>, country_code <chr>,
## # admin_division_name <chr>, admin_division_population <dbl>,
## # longitude <dbl>, latitude <dbl># Handling missing data for "country_name"
globalLandslide_df <- globalLandslide_df[grep('^[A-Za-z]', globalLandslide_df$country_name),]5. Checking for outliers
To check if there are any outliers in our data-set, we will be using boxplots to easily view them.
## Warning in (function (z, notch = FALSE, width = NULL, varwidth = FALSE, : some
## notches went outside hinges ('box'): maybe set notch=FALSE
By looking at the above plot, we see that although the death count because of the Landslides over all the years was below 1000, there was one where the fatality count was 5000.
It must have been an enormous disaster.
Hence, this solves our 1st Question: 5000 people were killed in the largest landslide disaster ever recorded
Data Transformation
1. Separating the Event_Date to Date and Time
We saw in the above section that one of the columns with missing value is “event_time”.
Let’s check how many cells of the “event_time” are empty.
sum(is.na(globalLandslide_df$event_time))## [1] 9471There is no data in the event_time variable as 11033 cells are empty.
This transformation step will get the time value from the “event_date” variable.
# Checking the values of the event_date column
head(globalLandslide_df$event_date, 5)## [1] "08/01/2008 12:00:00 AM" "01/02/2009 02:00:00 AM" "01/19/2007 12:00:00 AM"
## [4] "07/31/2009 12:00:00 AM" "10/16/2010 12:00:00 PM"By just checking the 5 records of the “event_date”, we can say that it also contains the time. So, we will be using the Solution-2 in this Data Transformation Step
Let’s separate the time from the “event_date” and store it into the “event_time” column
# Splitting the event_date by the first space
ev_dates <- as.data.frame(str_split_fixed(globalLandslide_df$event_date, " ", 2))
colnames(ev_dates) <- c("DATE", "TIME")
head(ev_dates, 10)## DATE TIME
## 1 08/01/2008 12:00:00 AM
## 2 01/02/2009 02:00:00 AM
## 3 01/19/2007 12:00:00 AM
## 4 07/31/2009 12:00:00 AM
## 5 10/16/2010 12:00:00 PM
## 6 02/16/2012 12:00:00 AM
## 7 03/30/2012 12:00:00 AM
## 8 09/02/2007 12:00:00 AM
## 9 09/05/2007 12:00:00 AM
## 10 11/01/2008 12:00:00 AMNow storing the “TIME” column of y to our “event_time” and “DATE” column of y to our event_date
globalLandslide_df$event_date <- ev_dates$DATE
globalLandslide_df$event_time <- ev_dates$TIMENow, let’s check the missing data in the event_time
sum(is.na(globalLandslide_df$event_time))## [1] 0In the above step, we performed Data-Transformation for the “event_date” and “event_time” variable.
# Removing the landslide size which is 'unknown'
globalLandslide_df <- globalLandslide_df[!grepl('unknown',globalLandslide_df$landslide_size),]Visual and Descriptive Analysis
1. Analyzing single Categorical Variables
Now, let us start by analyzing our 2 Categorical Variables - “landslide_size” and “landslide_category.”
The below query returns a table of frequency occurrences of data in each “landslide_size” category
table(globalLandslide_df$landslide_size)##
## large medium small very_large
## 632 6039 1908 86We can see that our data is not equally distributed among the different sizes of landslides by the output. And there are nine landslides whose size is not determined in our dataset.
Let’s see this in a Visual format.
# Create a bar graph of wool observations
ggplot(data = globalLandslide_df) +
geom_bar(mapping = aes(x = landslide_size))
The above bar graph clarifies that the highest number of landslides that occurred so far is medium-sized. The size of the maximum occurring landslides in our dataset is ** 2 categories of Wool are equally distributed in our dataset. There are nine landslides with no measure in our dataset.
This answers Question 2. No, the landslide sizes are not equally distributed in our dataset.
In the above plot, we can see that the number of very_large landslides is minimal. So, we can easily merge the large and very_large landslides as “large.”
globalLandslide_df$landslide_size[globalLandslide_df$landslide_size == "very_large"] <- "large"2. Analyzing the spread of landslides based on the category
Now, let us plot the landslide based on the category
# Create a bar graph of wool observations
p <- ggplot(data = globalLandslide_df) +
geom_bar(mapping = aes(x = landslide_category))
p + theme(axis.text.x = element_text(color="#993333",
size=10, angle=45),
axis.text.y = element_text(color="#993333",
size=10, angle=45))
From the above plot, we can see that most landslides are categorized as landslides. We also have significant mudslides and rock-fall types of landslides in our dataset.
3. Analyzing the number of injuries
# Create a boxplot of breaks
boxplot(
x = globalLandslide_df$injury_count,
xlab = "Number of Injuries",
horizontal = TRUE,
main = "BoxPlot of Injury Count recorded on the Landslide dataset"
)
The above Box Plot shows the spread of the injury count across our data frame
3. Plotting only those coutries with total injury more than 50 on a Pie Plot
landslide_distribution <- globalLandslide_df %>%
group_by(country_name) %>%
summarise(sum_injuries = sum(injury_count)) %>%
arrange(-sum_injuries)
landslide_distribution## # A tibble: 141 x 2
## country_name sum_injuries
## <chr> <dbl>
## 1 Guatemala 408
## 2 China 318
## 3 Peru 277
## 4 Nepal 257
## 5 Myanmar [Burma] 224
## 6 India 217
## 7 Bangladesh 170
## 8 Philippines 138
## 9 Indonesia 129
## 10 Brazil 103
## # ... with 131 more rowsgraph_data <- landslide_distribution[apply(landslide_distribution[,-1], 1, function(x) !all(x<=50)),]
x <- c(graph_data$sum_injuries)
yy <- c(graph_data$country_name)
pie(x, yy, main = "Country pie chart with more than 50 Injuries", edges = 10) 
The above plot shows the countries which have more than 50 injury_count in any of the landslides.
Hypothesis testing
In this section, we will move on to finding further answers to the hypothesis question listed above.
Now, let’s find out about the variables with character datatypes
data_char<-globalLandslide_df %>% dplyr::select(where(is.character))
for(i in colnames(data_char)){
print(unique(data_char[i]))
}## # A tibble: 3,319 x 1
## source_name
## <chr>
## 1 AGU
## 2 Oregonian
## 3 CBS News
## 4 Reuters
## 5 The Freeman
## 6 BusinessWorld Online
## 7 The Spokesman-Review
## 8 Crónica Diaria
## 9 MagicValley.com
## 10 UPI
## # ... with 3,309 more rows
## # A tibble: 2,670 x 1
## event_date
## <chr>
## 1 08/01/2008
## 2 01/02/2009
## 3 01/19/2007
## 4 07/31/2009
## 5 10/16/2010
## 6 02/16/2012
## 7 03/30/2012
## 8 09/02/2007
## 9 09/05/2007
## 10 11/01/2008
## # ... with 2,660 more rows
## # A tibble: 269 x 1
## event_time
## <chr>
## 1 12:00:00 AM
## 2 02:00:00 AM
## 3 12:00:00 PM
## 4 10:24:00 PM
## 5 08:30:00 PM
## 6 01:41:00 AM
## 7 01:00:00 AM
## 8 06:00:00 AM
## 9 08:50:00 AM
## 10 01:00:00 PM
## # ... with 259 more rows
## # A tibble: 8,406 x 1
## event_title
## <chr>
## 1 "Sigou Village, Loufan County, Shanxi Province"
## 2 "Lake Oswego, Oregon"
## 3 "San Ramon district, 195 miles northeast of the capital, Lima, "
## 4 "Dailekh district"
## 5 "sitio Bakilid in barangay Lahug"
## 6 "Paguite, Abuyog, Leyte"
## 7 "Pend Oreille County, State Route 20 near Usk, OR"
## 8 "3 killed in Acapulco"
## 9 "Warm Springs Road, Idaho"
## 10 "Lincang City, Yunnan, Yunnan-Tibet No. 214 highway."
## # ... with 8,396 more rows
## # A tibble: 7,879 x 1
## event_description
## <chr>
## 1 "occurred early in morning, 11 villagers buried in 7 houses"
## 2 "Hours of heavy rain are to blame for an overnight mudslide in Lake Oswego. "
## 3 "(CBS/AP) At least 10 people died and as many as 80 were still missing Wedne~
## 4 "One person was killed in Dailekh district, police said."
## 5 "Another landslide in sitio Bakilid in barangay Lahug also left two families~
## 6 "Thursday’s landslides were noted in Barangays Burubudan, Tadoc and Paguit~
## 7 "In Pend Oreille County, a mudslide on State Route 20 near Usk forced Washin~
## 8 "3 killed, including 2 children when rocks fell on their homes"
## 9 "5 feet deep mud, Hotshot crew was trapped while cleaning debris from fire"
## 10 "The report said heavy rainfall since Oct. 24 had hit 13 cities and counties~
## # ... with 7,869 more rows
## # A tibble: 8,322 x 1
## location_description
## <chr>
## 1 "Sigou Village, Loufan County, Shanxi Province"
## 2 "Lake Oswego, Oregon"
## 3 "San Ramon district, 195 miles northeast of the capital, Lima, "
## 4 "Dailekh district"
## 5 "sitio Bakilid in barangay Lahug"
## 6 "Paguite, Abuyog, Leyte"
## 7 "Pend Oreille County, State Route 20 near Usk, OR"
## 8 "calle Granjas, Ampliación Miguel de la Madrid, Acapulco"
## 9 "Warm Springs Road, Idaho"
## 10 "Lincang City, Yunnan, Yunnan-Tibet No. 214 highway."
## # ... with 8,312 more rows
## # A tibble: 9 x 1
## location_accuracy
## <chr>
## 1 unknown
## 2 5km
## 3 10km
## 4 25km
## 5 1km
## 6 50km
## 7 exact
## 8 100km
## 9 250km
## # A tibble: 13 x 1
## landslide_category
## <chr>
## 1 landslide
## 2 mudslide
## 3 complex
## 4 rock_fall
## 5 debris_flow
## 6 riverbank_collapse
## 7 unknown
## 8 lahar
## 9 other
## 10 snow_avalanche
## 11 creep
## 12 earth_flow
## 13 translational_slide
## # A tibble: 16 x 1
## landslide_trigger
## <chr>
## 1 rain
## 2 downpour
## 3 monsoon
## 4 tropical_cyclone
## 5 unknown
## 6 continuous_rain
## 7 mining
## 8 no_apparent_trigger
## 9 snowfall_snowmelt
## 10 flooding
## 11 dam_embankment_collapse
## 12 earthquake
## 13 construction
## 14 other
## 15 volcano
## 16 freeze_thaw
## # A tibble: 3 x 1
## landslide_size
## <chr>
## 1 large
## 2 small
## 3 medium
## # A tibble: 201 x 1
## storm_name
## <chr>
## 1 ""
## 2 "Supertyphoon Juan (Megi)"
## 3 "Tropical Storm Henrietta"
## 4 "Hurricane Dora"
## 5 "Agaton"
## 6 "Typhoon Nina"
## 7 "Tropical Depression 16"
## 8 "Typhoon No. 2 and March 11th earthquake"
## 9 "Typhoon Nepartak"
## 10 "Tropical Storm Alma"
## # ... with 191 more rows
## # A tibble: 141 x 1
## country_name
## <chr>
## 1 China
## 2 United States
## 3 Peru
## 4 Nepal
## 5 Philippines
## 6 Mexico
## 7 Algeria
## 8 Malaysia
## 9 Indonesia
## 10 Sierra Leone
## # ... with 131 more rows
## # A tibble: 140 x 1
## country_code
## <chr>
## 1 CN
## 2 US
## 3 PE
## 4 NP
## 5 PH
## 6 MX
## 7 DZ
## 8 MY
## 9 ID
## 10 SL
## # ... with 130 more rows
## # A tibble: 888 x 1
## admin_division_name
## <chr>
## 1 Shaanxi
## 2 Oregon
## 3 JunÃn
## 4 Mid Western
## 5 Central Visayas
## 6 Eastern Visayas
## 7 Washington
## 8 Sinaloa
## 9 Idaho
## 10 Yunnan
## # ... with 878 more rowsStudying correlation
In order to study the correlation in the dataset, let’s take the numerical variables of importance
numerical_data <- globalLandslide_df[, c('fatality_count', 'injury_count', 'longitude', 'latitude')] # Numerical variables
# Removing na values
numerical_data <- na.omit(numerical_data)library(corrplot)## corrplot 0.90 loadedlibrary(RColorBrewer)
corr <-cor(numerical_data)
corrplot(corr, type="upper", order="hclust",
col=brewer.pal(n=8, name="RdYlBu"))
The above plot solves answers our Question-3. There is a very minimum correlation between the fatality_count and injury_count. However, the only correlation we can find is between latitude and longitude.
Hypothesis 1: Is the mean value of fatality_count in one country equal to the mean value of fatality_count of another country with the same number of landslide
First, let’s take the frequency table of the various countries
country_tbl <- table(data_char$country_name)
country_tbl <- sort(country_tbl, decreasing = T)
country_tbl##
## United States India
## 2224 1261
## Philippines Nepal
## 669 479
## China Indonesia
## 425 350
## United Kingdom Brazil
## 225 214
## Canada Malaysia
## 173 166
## Pakistan Vietnam
## 141 116
## New Zealand Australia
## 106 105
## Colombia Mexico
## 101 86
## Guatemala Japan
## 82 82
## Thailand Costa Rica
## 77 76
## Sri Lanka Taiwan
## 75 66
## Trinidad and Tobago Bangladesh
## 65 58
## Peru Italy
## 58 55
## Kenya Uganda
## 53 45
## Panama Myanmar [Burma]
## 44 43
## Georgia Honduras
## 39 39
## Jamaica Bulgaria
## 37 35
## Fiji Ecuador
## 34 32
## Kyrgyzstan Nicaragua
## 32 31
## Ireland El Salvador
## 23 22
## France Haiti
## 22 22
## Norway Azerbaijan
## 22 21
## Papua New Guinea South Africa
## 21 21
## Turkey Venezuela
## 21 21
## Bhutan Tajikistan
## 20 20
## Switzerland Dominican Republic
## 19 17
## Dominica Afghanistan
## 16 15
## Nigeria Brunei
## 15 14
## Chile Spain
## 14 14
## Russia Bosnia and Herzegovina
## 13 12
## Bolivia South Korea
## 11 11
## Austria Lebanon
## 10 9
## Saint Lucia Argentina
## 8 7
## Ghana Ivory Coast
## 7 7
## Madagascar Puerto Rico
## 7 7
## Sierra Leone Yemen
## 7 7
## Iran Portugal
## 6 6
## Rwanda American Samoa
## 6 5
## Greece Saint Vincent and the Grenadines
## 5 5
## Saudi Arabia Serbia
## 5 5
## Solomon Islands Tanzania
## 5 5
## Armenia Cameroon
## 4 4
## Guinea Iceland
## 4 4
## Isle of Man Laos
## 4 4
## Macedonia North Korea
## 4 4
## Angola Cuba
## 3 3
## Democratic Republic of the Congo Germany
## 3 3
## Romania Ukraine
## 3 3
## Bermuda Croatia
## 2 2
## Czechia East Timor
## 2 2
## Ethiopia Grenada
## 2 2
## Hong Kong Israel
## 2 2
## Liberia Luxembourg
## 2 2
## Namibia Poland
## 2 2
## Slovakia U.S. Virgin Islands
## 2 2
## Vanuatu Albania
## 2 1
## Algeria Barbados
## 1 1
## Belize Burkina Faso
## 1 1
## Burundi Cambodia
## 1 1
## Czech Republic Egypt
## 1 1
## Gabon Guam
## 1 1
## Jersey Jordan
## 1 1
## Kazakhstan Malawi
## 1 1
## Mauritius Mongolia
## 1 1
## Montenegro Morocco
## 1 1
## Oman Paraguay
## 1 1
## Republic of the Congo Saint Kitts and Nevis
## 1 1
## Singapore Slovenia
## 1 1
## Sudan Swaziland
## 1 1
## United Arab Emirates Uzbekistan
## 1 1
## Zambia
## 1Above is the frequency count of how many times a country had a landslide recorded.
CASE-1
Looking at the above distribution, we see that the ‘Guatemala’ and ‘Japan’ had same number of landslides Now, let’s see if there is any relation between their fatality counts
data_globalLandslide_Japan <- filter(globalLandslide_df,country_name== "Japan")
data_globalLandslide_Guatemala <- filter(globalLandslide_df, country_name =="Guatemala")Now, let’s perform the T-Test
(mean(data_globalLandslide_Japan$fatality_count,na.rm=TRUE))## [1] 3.097222(mean(data_globalLandslide_Guatemala$fatality_count,na.rm=TRUE))## [1] 9.525641(var(data_globalLandslide_Japan$fatality_count,na.rm=TRUE))## [1] 99.52563(var(data_globalLandslide_Guatemala$fatality_count,na.rm=TRUE))## [1] 1789.707(t.test(data_globalLandslide_Guatemala$fatality_count,data_globalLandslide_Japan$fatality_count, alternative = "two.sided", var.equal = FALSE))##
## Welch Two Sample t-test
##
## data: data_globalLandslide_Guatemala$fatality_count and data_globalLandslide_Japan$fatality_count
## t = 1.3033, df = 86.218, p-value = 0.1959
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -3.376251 16.233089
## sample estimates:
## mean of x mean of y
## 9.525641 3.097222By looking at the mean of X and Y, we can say that although the number of landslides was the same, the mean of fatality count of these two countries is very different, and the alternative hypothesis is true
CASE-2
Now, let’s take another two countries with same number of landslides - Afghanistan and Nigeria with 15 landslides
data_globalLandslide_Afghanistan <- filter(globalLandslide_df,country_name== "Afghanistan")
data_globalLandslide_Nigeria <- filter(globalLandslide_df, country_name =="Nigeria")Now, let’s perform the T-Test on the injury_count
(mean(data_globalLandslide_Afghanistan$fatality_count,na.rm=TRUE))## [1] 191.1667(mean(data_globalLandslide_Nigeria$fatality_count,na.rm=TRUE))## [1] 2.272727(var(data_globalLandslide_Afghanistan$fatality_count,na.rm=TRUE))## [1] 362483.2(var(data_globalLandslide_Nigeria$fatality_count,na.rm=TRUE))## [1] 6.418182(t.test(data_globalLandslide_Afghanistan$fatality_count,data_globalLandslide_Nigeria$fatality_count, alternative = "two.sided", var.equal = FALSE))##
## Welch Two Sample t-test
##
## data: data_globalLandslide_Afghanistan$fatality_count and data_globalLandslide_Nigeria$fatality_count
## t = 1.0868, df = 11, p-value = 0.3004
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -193.6423 571.4302
## sample estimates:
## mean of x mean of y
## 191.166667 2.272727Here, as well there is a big difference on the fatality_count of the two countries And the alternative hypothesis is true
Fitting a Logistic Regression model to predict landslide_size based on Country name and fatality_count
sizeTbl <- table(data_char$landslide_size)
sizeTbl <- sort(sizeTbl, decreasing = T)
sizeTbl##
## medium small large
## 6039 1908 718As we have 3 categories in our landslide_size, we will be using “multinorm”
require(nnet)## Loading required package: nnetglobalLandslide_df$landslide_size <- as.factor(globalLandslide_df$landslide_size)
globalLandslide_df$landslide_size <- relevel(globalLandslide_df$landslide_size, ref = "small")
(test <- multinom(landslide_size ~ fatality_count+country_name, data = globalLandslide_df))## # weights: 408 (270 variable)
## initial value 8048.433627
## iter 10 value 5038.296265
## iter 20 value 4870.109640
## iter 30 value 4803.653120
## iter 40 value 4768.915852
## iter 50 value 4731.029628
## iter 60 value 4727.591801
## iter 70 value 4724.689782
## iter 80 value 4723.435549
## iter 90 value 4723.303577
## iter 100 value 4723.088146
## final value 4723.088146
## stopped after 100 iterations## Call:
## multinom(formula = landslide_size ~ fatality_count + country_name,
## data = globalLandslide_df)
##
## Coefficients:
## (Intercept) fatality_count country_nameAlbania country_nameAlgeria
## large -0.578098 0.2300878 -2.515561 -3.067708
## medium 1.543283 0.1469364 7.845234 8.521871
## country_nameAmerican Samoa country_nameAngola country_nameArgentina
## large -3.25849 -3.196976 12.59746
## medium 12.07988 11.332459 12.21304
## country_nameArmenia country_nameAustralia country_nameAustria
## large 0.4118777 -2.2002789 -3.294978
## medium -0.9465639 -0.9870048 12.970072
## country_nameAzerbaijan country_nameBangladesh country_nameBelize
## large 0.5367374 -1.756978 -15.46752
## medium 0.5124465 -1.428416 -25.56755
## country_nameBermuda country_nameBhutan country_nameBolivia
## large -15.449782 -13.055964 15.89792
## medium -1.543018 0.794617 14.38909
## country_nameBosnia and Herzegovina country_nameBrazil country_nameBrunei
## large -12.9279139 0.7411658 -16.508627
## medium 0.7996407 1.3354296 -0.479314
## country_nameBulgaria country_nameBurkina Faso country_nameBurundi
## large -1.788034 14.378477 -3.438488
## medium -0.937253 -6.785308 8.971612
## country_nameCambodia country_nameCameroon country_nameCanada
## large -2.792770 -3.014343 -1.549402
## medium 8.187121 10.134804 -1.306708
## country_nameChile country_nameChina country_nameColombia
## large -12.9914060 0.6695142 1.4672715
## medium 0.7178465 0.7860384 0.6235825
## country_nameCosta Rica country_nameCroatia country_nameCuba
## large -0.9082834 -3.014343 -3.196976
## medium -0.4974509 10.134804 11.332459
## country_nameCzech Republic country_nameCzechia
## large -2.515561 -15.449782
## medium 7.845234 -1.543018
## country_nameDemocratic Republic of the Congo country_nameDominica
## large 7.383624 -0.4178538
## medium 7.590342 -0.8044751
## country_nameDominican Republic country_nameEast Timor
## large -2.4087788 -3.765105
## medium -0.2519003 10.815946
## country_nameEcuador country_nameEgypt country_nameEl Salvador
## large -0.6601672 -5.894841 -0.8680328
## medium -0.4858372 12.308950 -0.5965775
## country_nameEthiopia country_nameFiji country_nameFrance
## large 8.082781 -0.7133941 -1.485874
## medium 8.497245 -0.8018322 -1.075178
## country_nameGabon country_nameGeorgia country_nameGermany
## large -2.792770 -0.200735808 -3.070858
## medium 8.187121 0.009777963 10.191661
## country_nameGhana country_nameGreece country_nameGrenada
## large -3.351408 12.13112 -18.29449
## medium 12.637225 11.12895 -25.82497
## country_nameGuam country_nameGuatemala country_nameGuinea
## large -15.46752 -0.7129694 12.32142
## medium -25.56755 0.6217765 11.06572
## country_nameHaiti country_nameHonduras country_nameHong Kong
## large -0.6918849 -1.737976 -3.12791
## medium -0.5143962 -0.869441 10.24737
## country_nameIceland country_nameIndia country_nameIndonesia
## large 1.270953 -0.6106655 1.220030
## medium -1.544228 0.1160878 1.126124
## country_nameIran country_nameIreland country_nameIsle of Man
## large -14.795593 -1.0594665 -15.449782
## medium -0.850085 -0.8716125 -1.543018
## country_nameIsrael country_nameItaly country_nameIvory Coast
## large -3.12791 0.5530436 -1.128185
## medium 10.24737 0.6898521 -0.907184
## country_nameJamaica country_nameJapan country_nameJersey
## large -16.2030949 -0.2702735 -15.46752
## medium 0.4955841 -0.2608066 -25.56755
## country_nameJordan country_nameKazakhstan country_nameKenya
## large 17.182869 -2.515561 0.92307215
## medium -9.921294 7.845234 0.04609626
## country_nameKyrgyzstan country_nameLaos country_nameLebanon
## large -2.43992 9.735334 -3.070224
## medium 15.24450 8.445189 13.313990
## country_nameLiberia country_nameLuxembourg country_nameMacedonia
## large -3.014343 -3.014343 -3.25849
## medium 10.134804 10.134804 12.07988
## country_nameMadagascar country_nameMalawi country_nameMalaysia
## large 11.62043 -2.515561 -0.6686410
## medium 11.53541 7.845234 -0.3115964
## country_nameMauritius country_nameMexico country_nameMongolia
## large -15.46752 0.6674792 18.30748
## medium -25.56755 0.7724658 -11.54534
## country_nameMontenegro country_nameMorocco country_nameMyanmar [Burma]
## large -2.515561 -2.976083 12.09529
## medium 7.845234 8.410691 11.71660
## country_nameNamibia country_nameNepal country_nameNew Zealand
## large -15.449782 0.2216214 -0.7733699
## medium -1.543018 0.6563372 -0.2068546
## country_nameNicaragua country_nameNigeria country_nameNorth Korea
## large -0.5550472 0.08686268 -3.014343
## medium -0.1612596 0.37172164 10.134804
## country_nameNorway country_namePakistan country_namePanama
## large 0.3240648 0.2468113 -1.4258222
## medium 1.2387180 0.4932025 -0.8757357
## country_namePapua New Guinea country_namePeru country_namePhilippines
## large 0.8863399 1.1068262 -0.2340531
## medium 0.6762315 0.8673936 0.4631328
## country_namePoland country_namePortugal country_namePuerto Rico
## large -3.014343 10.00997 -16.8816586
## medium 10.134804 10.34692 -0.8502976
## country_nameRomania country_nameRussia country_nameRwanda
## large -3.196976 0.4460222 10.95641
## medium 11.332459 -0.3597444 10.06126
## country_nameSaint Kitts and Nevis country_nameSaint Lucia
## large -15.46752 -3.323585
## medium -25.56755 12.982608
## country_nameSaint Vincent and the Grenadines country_nameSaudi Arabia
## large 12.38428 12.46179
## medium 11.73991 11.74287
## country_nameSerbia country_nameSierra Leone country_nameSlovakia
## large -14.3787353 -1.163464 -3.014343
## medium -0.4446624 -1.431597 10.134804
## country_nameSolomon Islands country_nameSouth Africa
## large -3.25849 -21.130129
## medium 12.07988 -1.729319
## country_nameSouth Korea country_nameSpain country_nameSri Lanka
## large 13.77458 -18.0733129 0.4103205
## medium 13.32137 -0.8737196 1.2120124
## country_nameSudan country_nameSwaziland country_nameSwitzerland
## large -3.067708 -2.608403 -0.8785222
## medium 8.521871 7.960248 -1.7192888
## country_nameTaiwan country_nameTajikistan country_nameTanzania
## large 1.296273 0.08543273 11.27755
## medium 1.442904 -0.49442824 10.92888
## country_nameThailand country_nameTrinidad and Tobago country_nameTurkey
## large -1.361164 -2.287774 -0.76323462
## medium -1.067582 -1.137821 0.03199308
## country_nameU.S. Virgin Islands country_nameUganda country_nameUkraine
## large -2.515561 12.47660 11.57391
## medium 7.845234 12.51994 10.20178
## country_nameUnited Kingdom country_nameUnited States
## large -1.744838 -2.352257
## medium -1.298424 -1.739923
## country_nameUzbekistan country_nameVanuatu country_nameVenezuela
## large -2.976083 10.455074 0.2809421
## medium 8.410691 8.458132 1.2415234
## country_nameVietnam country_nameYemen country_nameZambia
## large 0.8859336 12.46473 -3.344718
## medium 1.4174911 11.85391 8.857517
##
## Residual Deviance: 9446.176
## AIC: 9986.176The Residual Deviance here depicts how much the curve cannot fit and is very high.
In this case, our Logistic Regression failed. Hence, using the fatality_count and country_name, we were not able to predict the landslide_size as medium and large, keeping small as base
Summary of findings and questions for further analysis
We found the solution to Question-1(What was the landslide’s maximum death count so far?) We found the answer by plotting the fatality_count and looking at the maximum outlier. It could be an erroneous record, but the dataset can also have the correct number, and the death count could be 5000
We found out the solution to Question-2(Are the sizes of various landslides equally distributed in the dataset)
No, the dataset is highly skewed towards landslides’ “medium” size. The maximum values in the dataset were recorded for the “medium” landslide. Hence the various landslide sizes are not equally distributed in the dataset.We found out the solution to Question-3(What are the countries with more than 50 injured recorded in any landslide?) and used a pie-chart above to show the countries with more than 50 people wounded during any global landslide.
We found out the solution to Question-4(Is there any correlation between the numerical variable)
We only took the fatality_count, injury_count, latitude, and longitude to study the correlation between the numerical variables. However, we found no significant correlation between the location(depicted by latitude and longitude) and fatality_count or injury_count. We also found a very minimum correlation between the count of injured and demised people.For our Question-5(Perform hypothesis testing to see if the mean of the fatality_count of any two countries with the same number of landslides will be the same or not), we performed a T-test on two sets of countries with the same number of landslides. In both cases, we found that the mean value of fatality_count is no match. Even if those two countries have the same number of landslides recorded, one of them lost more lives than the other.
For our Question-5(Use Logistic Regression to predict the size of the landslide), we used logistic regression to predict the size of the landslide based on the country name and fatality_count. But our model failed. Hence, we could not predict the size of the landslide and could not build an efficient model using this dataset.
Conclusion
With this study on the GLC dataset, we could identify some of the relevant queries we could form using this global data. And we were also able to find solutions to some of them. The result was that the landslide size or the landslide count does not predict the number of deaths. After seeing the records, we also found that so many medium-size landslides cause more fatality after seeing the records. Unfortunately, we could not create a helpful prediction model of the landslide size based on the various countries.
Limitations
During this project, I realized I kept going back to the materials to clarify my understanding of the extensive dataset analysis. However, I also found that I lack my knowledge of R programming. There were some objectives for which I had to code many lines manually, but I’m sure there are many libraries of R which could make the work simple in one sentence. With this project, I also got practical exposure to understanding massive datasets. This GLC dataset can be used to create predictive models to identify potential landslide regions and their impact.