| library(HMM) |
| library(readr) |
| library(stringdist) |
| usernames <- c() |
| scenarios <- c() |
| filenames <- c() |
| hmmResults <- c() |
| files <- |
| list.files( |
| path = "~/answers", |
| pattern = "*.csv", |
| full.names = T, |
| recursive = TRUE |
| ) |
| for (f in files) |
|
| username <- regmatches(f, regexpr("-[A-Za-z0-9]+∖ ∖.", f)) |
| username <- sub("-", "", username) |
| username <- sub("∖∖.", "", username) |
| usernames <- c(usernames, username) |
| scenario <- regmatches(f, regexpr("-[1-3]-", f)) |
| scenario <- sub("-", "", scenario) |
| scenario <- sub("-", "", scenario) |
| scenarios <- c(scenarios, scenario) |
| filenames <- c(filenames, f) |
|
| for (i in 1:length(usernames)) |
|
| answerFile <- |
| read_csv(filenames[i], |
| col_names = FALSE, |
| col_types = cols_only(X3 = col_guess())) |
| answerFileX3 <- |
| as.POSIXct(answerFileX3, format = "%m/%d/%Y %H:%M:%S", tz = "UTC") |
| user <- |
| cert_r4_2_dataset[cert_r4_2_datasetuser == usernames[i], ] |
| m <- |
| match(answerFileX3, userdate) #match answer file dates to user dates |
| week <- |
| userweek[m[1:length(m)]] #week in which the attack ACTUALLY occurred |
| allWeeks <- |
| split(cert_r4_2_dataset[cert_r4_2_datasetuser %in% usernames[i], ]activity, |
| cert_r4_2_dataset[cert_r4_2_datasetuser %in% usernames[i], ]week) |
| #Filter dataset to only include data relevent to chosen user. |
| indx <- |
| sapply(allWeeks, length) #Convert the allWeeks variable into DataFrame. |
| res <- |
| as.data.frame(do.call(cbind, lapply(allWeeks, length<-, max(indx)))) |
| ############## HMM Phase ######################## |
| hmm = initHMM(c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), c(1, 2, 3, 4, 5, 6, 7)) |
| #Initiate a 10 state HMM with 7 labels (which represent activities of user.) |
| model = baumWelch( |
| hmm, |
| na.omit(unlist(res[1:5])), |
| maxIterations = 20, |
| pseudoCount = 0.1, |
| delta = 0.01 |
| ) #Train our model with the first 5 weeks of user activity. |
| vector = c() |
| for (i in 6:length(res)) |
| #For the remaining weeks of activity... |
| |
| #What is the probability of a given observed sequence with respect to our model? |
| logForwardProbabilities = forward(modelhmm, na.omit(unlist(res[i]))) #... calculate the |
| probability of week i occurring against model... |
| like <- ((logForwardProbabilities)) |
| lenthOfLike <- (length(like) / 10) |
| answer <- sum(like[, lenthOfLike]) |
| vector[i - 5] <- |
| answer #... store result of probability in vector... |
| print(i) #Prints the current week to so we can see progress of computations in the console. |
| model = baumWelch( |
| modelhmm, |
| na.omit(unlist(res[1:i])), |
| maxIterations = 20, |
| pseudoCount = 0.1, |
| delta = 0.01 |
| ) #... and update model with week i. |
| |
| ########## Find Lowest Probability ############# |
| probability = 0 |
| for (result in vector) |
| |
| if ((result) < probability) |
| |
| probability = result |
| |
| |
| hmmWeek = match(probability, vector) + 5 #Find week which lowest probability occurred in. Offet is |
| +5 since our results start at week 6. |
| if (hmmWeek %in% week) |
| |
| hmmResults <- c(hmmResults, hmmWeek) |
| |
| else |
| |
| hmmResults <- c(hmmResults, "FALSE") |
| |
| |
| fullResults <- |
| data.frame(usernames, scenarios, filenames, hmmResults) |
