Kickstarter Projects

Description of the dataset

Kickstarter Projects dataset contains data about more than 300000 initiatives from the Kickstarter crowdfunding platform. Each entry describes one fundraiser launched in the platform - its identification, categorisation, timeframe, financial results and goals.

Discussion of data mining goals and success criteria

Using given values of variables we may be able to assess the expected result of the project - out of five possibilities: Failed, Live, Successful, Canceled or Suspended - based on type of project (category, main category), country, amount of money set as goal, number of people who support the project and amount of money they pledge.

Characteristics of the dataset

All the data come from Kickstarter Platform and are collected in two .csv files - for two time points: the first one reflects the state of the platform content in December 2016 and the second in January 2018. In the prior set information about 321616 unique projects can be found, whilst in the latter one this number reaches 375765.

Description of the attributes

Data overwiev

Project categories and states

One of the basic characteristics of each project is a classification into theme category and subcategory. The other feature is a state of the project - that will be a subject of our interest in further research and predictions. The graph bellow presents a joint number of projects assigned to each of main categories together with representation of projects’ states.

For further analysis we will use only projects that provide the most relevant information - those that were historical in a moment of downloading data from the platform (successful, failed or cancelled). Records belonging to remaining three state categories are not going to be taken into consideration, so they will be excluded from the data set.

The following graph, presents proportions of successful, failed and cancelled projects within main theme categories. It is worth noticing that order of categories is indicated by descending number of projects within them. (Same as in one of previous graphs: “Number and state structure of projects by category”). Percentage of successful projects seems to be unrelated to overall number of projects. In particular, “Dance”, that is a category with the lowest number of registered projects, is also a category with the greater percentage of successes.

Which project categories gain the most backers?

Having plotted the boxplots of log of backers attracted by project categories, we can conclude which topics attract more sponsors. What we observe is that more people are eager too fund “Games”, “Design” and “Comics”, while “Crafts” and “Journalism” are the least backers-attracting.

Another interesting information is how the number of registered projects hanged in time:

Geographical diversification of projects

Kickstarter platform is dominated by projects created in United States, with a quite significant representation of undertakings established in Great Britain;

As it was stated previously, majority of projects originated from US. However, we decided to check whether the country could be influential for a project success. Graphs in this section allow to observe countries’ success and failure rates in particular years. It can be also seen when the country appeared in the projects registry.

Data quality

Outliers

To look at outliers, we will check only two columns - usd_pledged_real and usd_goal_real, because we must ensure that the values are converted to the same currency to be able to compare them.

When we look at the boxplot of first variable (usd_goal_real) we can see that it is very highly influenced by very high observations. Looking at number of elements that fall within each quartile, we can see that almost 100 000 lies below first quartile (25%) and above third (75%). Hence this are the observations visible aside the boxplot. Only the same number (200 000) observations fall within 25%-75% of data.
When identifying outliers using R function identify_outliers, we can see that it returns around 45 000 observations with mean goal in USD of 322 148.

	Value_of_quartile	Num_of_observations
25%	2000	98786
50%	5500	102437
75%	15500	94586
100%	166361391	94767

Num_of_outliers	Mean_outlier_value
45508	322148

The second numeric variable, usd_pledged_real also shows huge number of outliers (around 50 000).

	Value_of_quartile	Num_of_observations
25%	31.00	95071
50%	624.33	94697
75%	4050.00	94652
100%	20338986.27	94679

Num_of_outliers	Mean_outlier_value
50578	58253.41

k2018 <- read.csv(file = "ks-projects-201801.csv", sep = ",", dec = ".")
k2 <- k2018

stopifnot(is.numeric(k2$goal))
stopifnot(is.numeric(k2$pledged))
stopifnot(is.numeric(k2$backers))
stopifnot(is.numeric(k2$usd.pledged))
stopifnot(is.numeric(k2$usd_pledged_real))
stopifnot(is.numeric(k2$usd_goal_real))

stopifnot(k2$goal>0)
stopifnot(k2$pledged>=0)
stopifnot(k2$backers>=0)
stopifnot(na.omit(k2$usd.pledged)>=0)
stopifnot(k2$usd_pledged_real>=0)
stopifnot(k2$usd_goal_real>=0)

print('The integirty is preserved')

## [1] "The integirty is preserved"