Exploratory data analysis (EDA) is a mission critical task underpinning the predominance of detection development and preparation for cybersecurity-centric machine learning. There are a number of actions that analysts can take to better understand a particular data set and ready it for more robust utilization. In the spirit of toolsmith, and celebration of this being the 150th issue since toolsmith’s inception in late 2006, consider what follows a collection of tools for your security data analytics tool kit.
The University of Maryland’s Center for International and Security Studies (CISSM) Cyber Attacks Database is an ideal candidate for experimental exploration. Per the dataset description, the database “brings together open-source information surrounding a range of publicly acknowledged cyber events on private and public organizations. Events from 2014 through present have been coded to standardize information on threat actor, threat actor country, motive, target, end effects, industry, and country of impact. Source links to the news source are also provided” (Harry & Gallagher, 2018). I asked the project’s principal investigators for data export access as the default UI content is not suited to raw ingestion. The resulting cissm-export.csv (through MAR 2023) file, and all of the code that follows, as well as a Jupyter/Colab notebook for convenient experimentation, are available to you via my CISSM-EDA repository.
We begin with with loading the necessary libraries, data ingestion, data frame construction (tibble), and tsibble creation, a time series tibble. Our series of experiments require correlationfunnel, devtools, forecast, fpp2, CGPfunctions, ggpubr, janitor, tidyverse, tsibble, TTR, and vtree. The following snippet installs packages only if needed. Note that we’re installing dataxray from my fork in order to take advantage of an update I made to the report_xray() function. This update enables the results of a dataxray report to render in a browser automatically, particularly useful when calling the function from a Jupyter/Colab notebook.
my_packages <- c("correlationfunnel", "devtools", "forecast", "fpp2", "CGPfunctions", "ggpubr", "janitor", "tidyverse", "tsibble", "TTR", "vtree") # Specify your packages not_installed <- my_packages[!(my_packages %in% installed.packages()[ , "Package"])] # Extract packages to be installed if(length(not_installed)) install.packages(not_installed) # Install packages devtools::install_github("holisticinfosec/dataxray") # Install dataxray
With packages installed, we call libraries and build important components for our exercises. Comments are inline for each step.
# Attach the requisite packages library(dataxray) library(forecast) library(fpp2) library(CGPfunctions) library(ggpubr) library(janitor) library(tidyverse) library(tsibble) library(TTR) library(vtree) # ingest the CISSM CAD data as a data frame df <- read_csv("CISSM-export.csv", show_col_types = FALSE) # shrink the data set to include only event dates and event types evtType <- tabyl(df, evtDate, event_type) # convert the reduced data frame to a tibble df1 <- as_tibble(evtType) # create an all events tsibble df1 |> mutate(evtDate = yearmonth(evtDate)) |> as_tsibble(index = evtDate) -> AllEvents # create disruptive events tsibble AllEvents |> select(evtDate,Disruptive) -> disruptive # create exploitative events tsibble AllEvents |> select(evtDate,Exploitative) -> exploitative
The disruptive and exploitative variables will be used later as we model, forecast, and plot the results.
We’ll take advantage of the df variable for dataxray, janitor, CGPfunctions, and vtree.
dataxray is an interactive table interface for data summaries. It provides an excellent, interactive first look at the data set. I set a specific working directory in my script, set yours as you see fit.
df %>% make_xray() %>% view_xray() df %>% report_xray(data_name = 'CISSM', study = 'ggplot2')
The result is a columnar view of the CISSM Cyber Attacks Database with variables, observations including missing and distinct data, and interactive figures per variable. The view_xray() function creates an RStudio IDE Viewer pane while report_xray() generates the same result as RMD and HTML files written to your local directory. I’ve hosted the HTML version here if you’d like to interact with it without having to run the analyses yourself.
We learn quickly that the data include 10642 observations and 13 variables.
Of 10642 observations, there are 109 months of data, with 973 unique actors, six actor types, nine motives, four event types and plethora others. Hovering over the interactive figures we discover that motives include espionage, financial, industrial-espionage, personal attack, political-espionage, protest, protest-financial, sabotage, and undetermined. We also learn that event types include disruptive, exploitative, mixed, and undetermined. More on this next where we use janitor to manipulate this data further. dataxray is a great opening salvo in your analysis attack, fulfilling descriptive statistics capabilities admirably.
janitor, CGPfunctions, and vtree
While looking for improved methods to count by group in R I discovered janitor via Sharon’s Infoworld article including “quick and easy ways to count by groups in R, including reports as data frames, graphics, and ggplot graphs” (Machlis, 2020).
janitor includes simple little tools for “examining and cleaning dirty data, including functions that format ugly data.frame column names, isolate duplicate records for further study, and provide quick one- and two-variable tabulations (i.e., frequency tables and crosstabs) that improve on the base R function table().”
The CISSM CAD data sets up perfectly for a variety of table views with counts and/or percentages. While base R’s table() and dplyr’s count() are perfectly useful, a little enrichment never hurt anyone.
As you can see, table and count perform perfectly well.
>table(df$event_type) Disruptive Exploitative Mixed Undetermined 3462 5561 1525 94 >df %>% + count(event_type) event_type n 1 Disruptive 3462 2 Exploitative 5561 3 Mixed 1525 4 Undetermined 94
What table and count don’t necessarily provide are the aforementioned enrichment, but tabyl and janitor’s adorn functions fill that void nicely.
First, basic tabyl use yields quick results as seen in the partial output snippet below.
>tabyl(df, event_type, motive) event_type Espionage Financial Industrial-Espionage Personal Attack Political-Espionage Protest Protest,Financial Sabotage Undetermined Disruptive 0 1335 0 27 6 643 0 240 1211 Exploitative 25 3077 86 64 497 282 0 42 1488 Mixed 0 1361 2 2 14 41 1 16 88 Undetermined 0 57 0 0 2 2 0 1 32
If we chose to adorn our results however the result is all the more beneficial.
>tabyl(df, event_type, motive) %>% + adorn_percentages("col") %>% + adorn_pct_formatting(digits = 1) event_type Espionage Financial Industrial-Espionage Personal Attack Political-Espionage Protest Protest,Financial Sabotage Undetermined Disruptive 0.0% 22.9% 0.0% 29.0% 1.2% 66.4% 0.0% 80.3% 43.0% Exploitative 100.0% 52.8% 97.7% 68.8% 95.8% 29.1% 0.0% 14.0% 52.8% Mixed 0.0% 23.3% 2.3% 2.2% 2.7% 4.2% 100.0% 5.4% 3.1% Undetermined 0.0% 1.0% 0.0% 0.0% 0.4% 0.2% 0.0% 0.3% 1.1%
Continuing our journey with event types and motives, a similar construct can be visualized using CGPfunctions.
Per the CGPfunctions description, it’s a package that includes miscellaneous functions useful for teaching statistics as well as actually practicing the art. They typically are not new methods but rather wrappers around either base R or other packages. In this case, we’ll use PlotXTabs2 which wraps around ggplot2 to provide bivariate bar charts for categorical and ordinal data, specifically the event_type and motive variables.
PlotXTabs2(df, event_type, motive, title = "Event Type by Motive")
PlotXTabs2 stands out as it offers a summary of key frequentist and bayesian information as a subtitle (can be suppressed) as well as a plethora of formatting options courtesy of ggstatsplot (Powell, 2020). Noteworthy in the resulting plot is the fact that across all event types (disruptive, exploitative, mixed, undetermined) the predominant motive was financial. Not surprising, but noteworthy. Regardless, PlotXTabs2 provides an incredibly useful visualization of the CAD dataset with specific variables.
Finally, we conclude this section with use of the vtree package. vtree, or variable trees, displays information about nested subsets of a data frame, in which the subsetting is defined by the values of categorical variables. This is again a perfect option for our subsetted CISSM CAD data. First, we generate a simple vtree based only on event type.
The result includes a breakdown of event types from the total of 10642 events, including counts and percentage. If you prefer a view without counts you can exclude them as follows; in this case we do so with motives. Note that vtree is incredibly rich with argument features, tailor your visualizations to your liking. Call
help("vtree") to learn more.
vtree(df, "motive", showcount = FALSE)
Finally, we join event types and motives, and pivot our tree to be read vertically.
vtree(df, c("event_type", "motive"), showcount = FALSE, horiz = FALSE)
The more data you join, the more unwieldy the tree becomes, but nonetheless an intriguing view when zoomed appropriately.
Next, we model the CISSM CAD data for time series forecasting with three well known methods selected for performance with the CISSM CAD dataset. These include naive, SES, and ARIMA.
Naive forecasting uses the most recent observation as the forecast for the next observation.
Simple exponential smoothing (SES) is the method of time series forecasting used with univariate data with no trend and no seasonal pattern.
Autoregressive integrated moving average, or ARIMA, is a statistical analysis model that uses time series data to either better understand the data set or to predict future trends.
Before we initiate the forecasts we use each of the methods to determine which one performs best. For brevity here, we’ll only run the models and plots on exploitative data, but the forecasts_CISSM.R script and the Jupyter/Colab notebook in the GitHub repo run all models and plots with disruptive and exploitative data. Note again that these are subsets of a much broader dataset represented by CISSM CAD. Explore further to your liking.
Important to the modeling that follows, note the evaluation metrics RMSE and MAE. Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) are used to evaluate regression models, tell us how accurate our predictions are, and the amount of deviation from actual values (Acharya, 2021). In essence, the lower the score, the better the performance.
> naive_model_exploitative <- naive(exploitative, h = 12) > summary(naive_model_exploitative) Forecast method: Naive method Model Information: Call: naive(y = exploitative, h = 12) Residual sd: 20.5334 Error measures: ME RMSE MAE MPE MAPE MASE ACF1 Training set -0.06481481 20.5334 15.41667 -Inf Inf 0.6228308 -0.3349566 Forecasts: Point Forecast Lo 80 Hi 80 Lo 95 Hi 95 Feb 2023 23 -3.314606 49.31461 -17.24472 63.24472 Mar 2023 23 -14.214473 60.21447 -33.91463 79.91463 Apr 2023 23 -22.578235 68.57824 -46.70590 92.70590 May 2023 23 -29.629213 75.62921 -57.48944 103.48944 Jun 2023 23 -35.841249 81.84125 -66.98992 112.98992 Jul 2023 23 -41.457358 87.45736 -75.57902 121.57902 #snipped
The naive model yields an RMSE score of 20.5 and an MAE score 15.4.
> ses_model_exploitative <- ses(exploitative$Exploitative, h = 12) # RMSE = 18.8, MAE = 13.9 > summary(ses_model_exploitative) Forecast method: Simple exponential smoothing Model Information: Simple exponential smoothing Call: ses(y = exploitative$Exploitative, h = 12) Smoothing parameters: alpha = 0.529 Initial states: l = 32.3171 sigma: 19.0189 AIC AICc BIC 1157.442 1157.671 1165.517 Error measures: ME RMSE MAE MPE MAPE MASE ACF1 Training set 0.2068036 18.84358 13.92841 -Inf Inf 0.9034647 0.03522811 Forecasts: Point Forecast Lo 80 Hi 80 Lo 95 Hi 95 110 44.24184 19.8681717 68.61551 6.965531 81.51815 111 44.24184 16.6677635 71.81592 2.070929 86.41275 112 44.24184 13.8020028 74.68168 -2.311874 90.79555 113 44.24184 11.1837445 77.29994 -6.316154 94.79983 114 44.24184 8.7581582 79.72552 -10.025768 98.50945 115 44.24184 6.4880896 81.99559 -13.497539 101.98122 #snipped
The SES model yields an RMSE score of 18.8 and an MAE score of 13.9.
> arima_model_exploitative <- auto.arima(exploitative) > summary(arima_model_exploitative) Series: exploitative ARIMA(2,1,1) Coefficients: ar1 ar2 ma1 0.4744 0.1671 -0.9653 s.e. 0.1021 0.1007 0.0323 sigma^2 = 343.6: log likelihood = -467.68 AIC=943.35 AICc=943.74 BIC=954.08 Training set error measures: ME RMSE MAE MPE MAPE MASE ACF1 Training set 2.152378 18.19375 13.5603 -Inf Inf 0.5478338 -0.01933472
Finally, the ARIMA model yields an RMSE score of 18.2 and an MAE score of 13.6. Ultimately, by a small margin, the ARIMA model is most likely to provide the best forecast. Next, we forecast and plot the results. Given that ARIMA is most reliable under these circumstances, we’ll focus on visualizing ARIMA results; you can experiment with naive and SES plots on your own via the scripts or the notebook.
Forecasts & Plots
Generating a plot of AllEvents is as easy as:
This is just as easy with disruptive or exploitative events exclusively with the likes of
To forecast the exploitative ARIMA model in an individual plot, utilize:
forecast(arima_model_exploitative) %>% autoplot()
The light and dark areas correspond to the 95% and 80% confidence intervals (CI) respectively.
You can join multiple plots to compare outcomes side by side as follows.
naiveEXP = forecast(naive_model_exploitative) %>% autoplot() sesEXP = forecast(ses_model_exploitative) %>% autoplot() arimaEXP = forecast(arima_model_exploitative) %>% autoplot() multi.pageEXP <- ggarrange(naiveEXP, sesEXP, arimaEXP, nrow = 3, ncol = 1) multi.pageEXP
You may be wondering what ARIMA(2,1,1) refers to in our plots. A nonseasonal ARIMA model, which this is, is classified as an “ARIMA(p,d,q)” model, where: p is the number of autoregressive terms, d is the number of nonseasonal differences needed for stationarity, and q is the number of lagged forecast errors in the prediction equation. Therefore, in this case, (2,1,1) is p,d,q found by the auto.arima process indicating that we have two auto-regessive terms, one difference, and one moving average term in our series (Nau, 2020).
Hopefully, this effort has been useful and insightful for security analysts as well as fledgling data scientists in the security realm. It’s no surprise that I orient towards the practices of visualization; I have found all methods deployed here to be useful, effective, and durable for future use. It is my desire that you benefit similarly, and that this opens some doors for you, literally and figuratively.
Cheers…until next time.
Acharya, S. (2021, June 15). What are RMSE and Mae? Medium. Retrieved May 3, 2023, from https://towardsdatascience.com/what-are-rmse-and-mae-e405ce230383
Harry, C., & Gallagher, N. (2018). Classifying cyber events. Journal of Information Warfare, 17(3), 17-31.
Machlis, S. (2020, September 10). How to count by group in R. InfoWorld. Retrieved May 1, 2023, from https://www.infoworld.com/article/3573577/how-to-count-by-groups-in-r.html
Nau, R. (2020, August 8). Introduction to ARIMA models. Statistical forecasting: notes on regression and time series analysis. Retrieved May 4, 2023, from https://people.duke.edu/~rnau/411arim.htm#pdq
Powell, C. (2020, November 12). Using PlotXTabs2. The Comprehensive R Archive Network. Retrieved May 2, 2023, from https://cran.r-project.org/web/packages/CGPfunctions/vignettes/Using-PlotXTabs2.html