Traffic-Prediction-using-Bayesian-Network-Model

ITC - University of Twente, Netherlands (Master Thesis | 2020)

About

As part of my academic master thesis, submitted to the faculty of Geo-Information Science and Earth Observation, Earth Observation Department, ITC-University of Twente.

The entire work flow is categorised into various different sections:-

1. Area Selection

For this case study, I have choosen major cities in Netherlands. Namely, Amsterdam & Rotterdam.

Indicators	Amsterdam	Rotterdam
Area covered	193.5 Sq km	161.6 Sq km
No. of edges in network	26580	25387
Avg street length	102.58	97.7

2. Bayesian Network Model

The model was validated using k-fold cross validation criteria where in the bayesian network was designed based on IAMB & Hill climbing learning algorithms.

3. Literature study

Various research papers were studied to understand the complexities of the baysian networks. The detailed study papers can be found in the folder 'literature study'.

4. R-Scripts

• 4.1 Bayesian Network Model

library(readxl)
library(bnlearn)
data1 = RTM_2020_Cat_Agg
data2 = lapply(data1, as.factor)
data3 = data.frame(data2)
pdag = hc(data3)
pdag
plot(pdag)
dag = set.arc(pdag, from = "Vehicle_Cat", to = "Average_Intensity")
dag1 = set.arc(dag, from = "Time.of.the.day", to = "Average_Time")
dag2 = set.arc(dag1, from = "Time.of.the.day", to = "Average_Intensity")
dag3 = drop.arc(dag2, from = "Day.in.a.week", to = "Average_Speed")
dag4 = drop.arc(dag3, from = "Day.in.a.week", to = "Week.Day.Week.End")
dag5 = set.arc(dag4, from = "Day.in.a.week", to ="Average_Speed")
dag6 = set.arc(dag5, from = "Week.Day.Week.End", to = "Average_Time")
dag7 = set.arc(dag6, from = "Average_Time", to = "Average_Intensity")
dag8 = set.arc(dag7, from = "Average_Time", to = "Average_Speed")
dag9 = set.arc(dag8, from = "Average_Speed", to = "Average_Intensity")
dag10 = drop.arc(dag9, from = "Day.in.a.week", to = "Average_Intensity")
getOption("max.print")
dag10
plot(dag10)
score(dag10, data3, type = "bic")
score(dag10, data3, type ="aic")
score(dag10, data3, type="bde")
score(dag10, data3, type="k2")

fit = bn.fit(dag10, data3)
print(fit$Day.in.a.week)
print(fit$Week.Day.Week.End)
print(fit$Time.of.the.day)
print(fit$Average_Speed)
print(fit$Average_Time)
print(fit$Average_Intensity)
print(fit$Vehicle_Cat)
fit
cpquery(fit, event = (Average_Time =="High"), evidence = (Average_Speed=="High"))
cpquery(fit, event = (Average_Intensity =="High"), evidence = (Average_Speed=="Low" & Average_Time =="High"))
bn.cv(data3, bn ="iamb", k =5)

# Hill climbing Algorithm
data3 = data.frame(data2)
pdag1 = hc(data3)
pdag1
plot(pdag1)
pdag1 = set.arc(pdag1, from = "Vehicle_Cat", to = "Average_Intensity")
plot(pdag1)
score(pdag1, data3, type = "bic")
score(pdag1, data3, type ="aic")
score(pdag1, data3, type="bde")
score(pdag1, data3, type="k2")

• 4.2 Bayesian Prediction Model

library(readxl)
library(bnlearn)
library(forecast)
data1 = Time_Wise_2020_D1_Cont
data3 = data.frame(data1)
abc = data3[1:200, ]
def = data3[201:743, ]
pdag = hc(abc)
pdag
plot(pdag)
dag = set.arc(pdag, from = "Vehicle_Cat", to = "Average_Intensity")
dag2 = set.arc(dag, from = "Time.of.the.day", to = "Average_Intensity")
dag3 = drop.arc(dag2, from = "Day.in.a.week", to = "Average_Speed")
dag4 = drop.arc(dag3, from = "Day.in.a.week", to = "Week_Day_Week_End")
dag5 = set.arc(dag4, from = "Day.in.a.week", to ="Average_Speed")
dag6 = set.arc(dag5, from = "Week_Day_Week_End", to = "Average_Time")
dag7 = set.arc(dag6, from = "Average_Time", to = "Average_Intensity")
dag8 = set.arc(dag7, from = "Average_Time", to = "Average_Speed")
dag9 = set.arc(dag8, from = "Average_Speed", to = "Average_Intensity")
options(max.print=1000000)
dag9
plot(dag9)
fitttted = bn.fit(dag9, abc)
pred = predict(fitttted, "Average_Intensity", def)
#cbind(pred, def[, "Average_Intensity"])
cbind(predicted =pred, actual = def[, "Average_Intensity"])
accuracy(forecast(pred))
accuracy(f = pred, x = def[, "Average_Intensity"])
accuracy(forecast(def$Average_Intensity))
accuracy(forecast(abc$Average_Intensity))

5. Sensor Data

Traffic data was downlaoded from National dataware house agency, a repository for traffic related information: NDW, NL

6. ShapeFiles

For the developement of BN model, various vector tiles was stacked on top of eachother, the vector files were used to extract the street related information.

7. Street Network [Outputs]

• Amsterdam

• Rotterdam