Crash Count Analysis_ 1
July 19, 2015
Data calling
setwd("/Users/subasishdas1/Copy/Rpubs/rpubs/Crash_exmp1")
all_a <- read.csv("All_a_new.csv")
head(all_a) ## Read first few rows## Avg_Crashes ADT Segment_Length Median_Width Number_Lanes Shoulder_Width
## 1 0.00 2264 0.31 0 2 5
## 2 1.00 337 0.23 0 2 0
## 3 1.67 3601 0.31 0 2 0
## 4 0.67 3962 0.13 0 2 0
## 5 7.00 3237 0.17 0 2 0
## 6 0.00 701 0.09 0 2 0
## Lane_Width Class CBD Type
## 1 11 1 0 twolane
## 2 10 1 0 twolane
## 3 10 1 0 twolane
## 4 11 2 0 twolane
## 5 12 2 0 twolane
## 6 10 1 1 twolanedim(all_a) ### Rows, Columns## [1] 90 10names(all_a) ## Name of the columns## [1] "Avg_Crashes" "ADT" "Segment_Length" "Median_Width"
## [5] "Number_Lanes" "Shoulder_Width" "Lane_Width" "Class"
## [9] "CBD" "Type"table(all_a$Type)##
## fourlane Int twolane
## 30 30 30twolane <- subset(all_a, Type=="twolane") ## subsetting
fourlane <- subset(all_a, Type=="fourlane")
int <- subset(all_a, Type=="Int")Two-lane
library(e1071)
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Median_Width
+Number_Lanes +Shoulder_Width +Lane_Width +Class +CBD, data=twolane, cost=100,gamma=1)## Warning in svm.default(x, y, scale = scale, ..., na.action = na.action):
## Variable(s) 'Median_Width' and 'Number_Lanes' constant. Cannot scale data.### modification
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Shoulder_Width
+Lane_Width +Class +CBD, data=twolane, cost=100,gamma=1) ### Remove two redundant columns. Check warnings above.
svm.pred <- predict(svm.model, twolane[2:9])
### write.csv(svm.pred, "svm_twolane.csv")
pl <- cbind(obs= twolane$Avg_Crashes, pred=svm.pred, diff=svm.pred-twolane$Avg_Crashes)
pl <- data.frame(pl)
pl## obs pred diff
## 1 0.00 0.2178539 0.21785392
## 2 1.00 1.2173730 0.21737301
## 3 1.67 1.4513990 -0.21860099
## 4 0.67 0.8870490 0.21704898
## 5 7.00 6.7817412 -0.21825881
## 6 0.00 0.2181368 0.21813677
## 7 1.33 1.4414940 0.11149397
## 8 1.00 0.9406227 -0.05937731
## 9 2.33 2.5480234 0.21802342
## 10 10.00 9.7820490 -0.21795103
## 11 3.00 2.7816793 -0.21832069
## 12 1.67 1.6829701 0.01297007
## 13 2.67 2.4514955 -0.21850453
## 14 1.67 1.4521225 -0.21787748
## 15 2.00 1.7828937 -0.21710634
## 16 0.00 0.2173967 0.21739671
## 17 0.00 0.2188750 0.21887499
## 18 0.00 0.2188183 0.21881830
## 19 0.00 0.2179866 0.21798665
## 20 1.00 1.2191915 0.21919146
## 21 1.33 1.5478579 0.21785794
## 22 2.00 1.7818863 -0.21811370
## 23 0.00 0.2177289 0.21772893
## 24 3.67 3.4521887 -0.21781132
## 25 0.33 0.4951871 0.16518713
## 26 0.67 0.6911078 0.02110782
## 27 0.67 0.8875763 0.21757630
## 28 0.33 0.5483805 0.21838053
## 29 0.00 0.2178195 0.21781951
## 30 0.00 0.2190978 0.21909784plot(pl$obs, pl$pred)
## Comment: model is good.
print(svm.model)##
## Call:
## svm(formula = Avg_Crashes ~ ADT + Segment_Length + Shoulder_Width +
## Lane_Width + Class + CBD, data = twolane, cost = 100, gamma = 1)
##
##
## Parameters:
## SVM-Type: eps-regression
## SVM-Kernel: radial
## cost: 100
## gamma: 1
## epsilon: 0.1
##
##
## Number of Support Vectors: 25Four-lane
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Median_Width +Number_Lanes
+ Shoulder_Width +Lane_Width +Class +CBD, data=fourlane,cost=100,gamma=1)## Warning in svm.default(x, y, scale = scale, ..., na.action = na.action):
## Variable(s) 'Median_Width' and 'CBD' constant. Cannot scale data.### modification
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Number_Lanes
+ Shoulder_Width +Lane_Width +Class, data=fourlane,cost=100,gamma=1)### Remove two redundant columns. Check warnings above.
svm.pred <- predict(svm.model, fourlane[2:9])
print(svm.model)##
## Call:
## svm(formula = Avg_Crashes ~ ADT + Segment_Length + Number_Lanes +
## Shoulder_Width + Lane_Width + Class, data = fourlane, cost = 100,
## gamma = 1)
##
##
## Parameters:
## SVM-Type: eps-regression
## SVM-Kernel: radial
## cost: 100
## gamma: 1
## epsilon: 0.1
##
##
## Number of Support Vectors: 27pl <- cbind(obs= fourlane$Avg_Crashes, pred=svm.pred, diff=svm.pred-fourlane$Avg_Crashes)
pl <- data.frame(pl)
pl## obs pred diff
## 31 2.67 2.9422229 0.27222294
## 32 0.67 0.9422006 0.27220058
## 33 8.00 7.7291994 -0.27080057
## 34 0.00 -0.1859454 -0.18594539
## 35 0.00 0.2718133 0.27181326
## 36 0.67 0.9420076 0.27200758
## 37 1.00 1.2723890 0.27238903
## 38 1.00 1.2729382 0.27293822
## 39 6.33 6.0571761 -0.27282394
## 40 0.67 0.9409801 0.27098013
## 41 4.00 3.7269801 -0.27301992
## 42 5.67 5.9419802 0.27198020
## 43 3.67 3.3979686 -0.27203142
## 44 6.67 6.3979686 -0.27203144
## 45 1.33 1.6020274 0.27202745
## 46 9.00 8.7293041 -0.27069591
## 47 3.00 3.0804473 0.08044729
## 48 4.00 3.7287935 -0.27120652
## 49 3.67 4.8744353 1.20443526
## 50 0.67 0.9414899 0.27148991
## 51 0.33 0.6015183 0.27151834
## 52 0.33 0.6024319 0.27243186
## 53 1.33 1.0589257 -0.27107426
## 54 1.67 1.9411041 0.27110409
## 55 2.00 2.2714316 0.27143156
## 56 1.33 1.6010484 0.27104836
## 57 1.00 0.7267401 -0.27325994
## 58 2.33 2.4748522 0.14485218
## 59 5.33 5.0588991 -0.27110092
## 60 9.00 8.7281731 -0.27182688plot(pl$obs, pl$pred)
## Comment: model is good.
print(svm.model)##
## Call:
## svm(formula = Avg_Crashes ~ ADT + Segment_Length + Number_Lanes +
## Shoulder_Width + Lane_Width + Class, data = fourlane, cost = 100,
## gamma = 1)
##
##
## Parameters:
## SVM-Type: eps-regression
## SVM-Kernel: radial
## cost: 100
## gamma: 1
## epsilon: 0.1
##
##
## Number of Support Vectors: 27Interstate
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Median_Width +Number_Lanes
+ Shoulder_Width +Lane_Width +Class +CBD, data=int,cost=100,gamma=1)## Warning in svm.default(x, y, scale = scale, ..., na.action = na.action):
## Variable(s) 'Lane_Width' and 'Class' and 'CBD' constant. Cannot scale data.### modification
svm.model <- svm(Avg_Crashes~ADT +Segment_Length +Number_Lanes
+ Shoulder_Width, data=fourlane,cost=100,gamma=1)### Remove three redundant columns. Check warnings above.
svm.pred <- predict(svm.model, int[2:9])
print(svm.model)##
## Call:
## svm(formula = Avg_Crashes ~ ADT + Segment_Length + Number_Lanes +
## Shoulder_Width, data = fourlane, cost = 100, gamma = 1)
##
##
## Parameters:
## SVM-Type: eps-regression
## SVM-Kernel: radial
## cost: 100
## gamma: 1
## epsilon: 0.1
##
##
## Number of Support Vectors: 27### write.csv(svm.pred, "svm_int.csv")
pl <- cbind(obs= int$Avg_Crashes, pred=svm.pred, diff=svm.pred-int$Avg_Crashes)
pl <- data.frame(pl)
pl## obs pred diff
## 61 3.33 1.738589 -1.59141106
## 62 4.67 1.738553 -2.93144699
## 63 0.33 1.738553 1.40855301
## 64 6.33 1.738578 -4.59142182
## 65 26.67 1.738553 -24.93144699
## 66 6.00 1.738553 -4.26144699
## 67 0.67 1.738553 1.06855301
## 68 1.67 1.738553 0.06855301
## 69 0.00 1.738806 1.73880567
## 70 1.00 1.738701 0.73870096
## 71 2.33 1.738553 -0.59144699
## 72 1.00 1.738553 0.73855301
## 73 0.33 1.738553 1.40855321
## 74 0.67 1.738553 1.06855301
## 75 2.00 1.738596 -0.26140355
## 76 0.00 1.738553 1.73855301
## 77 0.00 1.738553 1.73855301
## 78 4.67 1.738553 -2.93144699
## 79 9.67 1.738557 -7.93144345
## 80 0.33 1.739053 1.40905337
## 81 1.00 1.738745 0.73874466
## 82 0.33 1.738553 1.40855301
## 83 2.00 1.738914 -0.26108575
## 84 3.33 1.738556 -1.59144391
## 85 0.00 1.738553 1.73855301
## 86 1.33 1.738930 0.40893043
## 87 11.00 1.738553 -9.26144699
## 88 0.33 1.738553 1.40855301
## 89 7.67 1.738933 -5.93106668
## 90 14.33 1.738553 -12.59144699plot(pl$obs, pl$pred)
## Comment: model is not good at all.
print(svm.model)##
## Call:
## svm(formula = Avg_Crashes ~ ADT + Segment_Length + Number_Lanes +
## Shoulder_Width, data = fourlane, cost = 100, gamma = 1)
##
##
## Parameters:
## SVM-Type: eps-regression
## SVM-Kernel: radial
## cost: 100
## gamma: 1
## epsilon: 0.1
##
##
## Number of Support Vectors: 27Conducted by: Subasish Das