Maintenance price vs Car Acceptability
Customer acceptability evaluation on Cars using Classification
Chitra V
October 18 2015
Introduction:
Data_set: Car evaluation This Car evaluation dataset is taken from UCI Machine learning repository derived from simple hierarchical decision model
Total no. of Observations: 1728
Input Variable:
- Buying price (vhigh, high, med, low)
- Price of the maintenance (vhigh, high, med, low)
- Number of doors (2, 3, 4, 5more)
- Persons capacity in terms of persons to carry (2, 4, more)
- Size of luggage boot (small, med, big)
- Estimated safety of the car (low, med, high)
Output Variable:
- Car acceptability (unacc, acc, good, vgood)
Car_Evaluation attributes and its types
## 'data.frame': 1728 obs. of 7 variables:
## $ buying : Factor w/ 4 levels "high","low","med",..: 4 4 4 4 4 4 4 4 4 4 ...
## $ maint : Factor w/ 4 levels "high","low","med",..: 4 4 4 4 4 4 4 4 4 4 ...
## $ doors : Factor w/ 4 levels "2","3","4","5more": 1 1 1 1 1 1 1 1 1 1 ...
## $ persons : Factor w/ 3 levels "2","4","more": 1 1 1 1 1 1 1 1 1 2 ...
## $ lug_boot : Factor w/ 3 levels "big","med","small": 3 3 3 2 2 2 1 1 1 3 ...
## $ safety : Factor w/ 3 levels "high","low","med": 2 3 1 2 3 1 2 3 1 2 ...
## $ Class.Value: Factor w/ 4 levels "acc","good","unacc",..: 3 3 3 3 3 3 3 3 3 3 ...All the variables are factor variables
Car acceptability grouping
Car evaluation based on various parameters
Seating capacity vs Car Acceptability
Conclusion: Seating capacity is an important factor for customers in accepting or rejecting a car
Car safety vs Car Acceptability
Conclusion: Safety is an important factor for Customers in accepting or rejecting a car
Low safety cars are not accepted by the customers
Buying price vs Car Acceptability
No of doors vs Car Acceptability
Luggage boot size vs Car Acceptability
Read the car evaluation data from car_data.csv
Split the data 50:50 for Training and testing purpose
Since the data set has multiclass output function, C50 rules, c50 tree and rpart tree algorithms have been used to predict the model
Algorithm 1: RPART TREE
Number & proportion of class values for unacceptable, acceptable, good and vgood
##
## unacc acc good vgood
## 1210 384 69 65##
## unacc acc good vgood
## 0.70023148 0.22222222 0.03993056 0.03761574——————————————————————————————-
Rpart 1 - Build a rpart tree with safety and persons variable (based on the stacked bars)
Variable Importance
## persons safety
## 85.35115 69.87636Create a rpart tree with safety and persons variables and compare with the original data
## Confusion Matrix and Statistics
##
## Reference
## Prediction unacc acc good vgood
## unacc 516 82 0 0
## acc 47 142 0 0
## good 9 29 0 0
## vgood 0 39 0 0
##
## Overall Statistics
##
## Accuracy : 0.7616
## 95% CI : (0.7317, 0.7896)
## No Information Rate : 0.662
## P-Value [Acc > NIR] : 1.242e-10
##
## Kappa : 0.4904
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: unacc Class: acc Class: good Class: vgood
## Sensitivity 0.9021 0.4863 NA NA
## Specificity 0.7192 0.9178 0.95602 0.95486
## Pos Pred Value 0.8629 0.7513 NA NA
## Neg Pred Value 0.7895 0.7778 NA NA
## Prevalence 0.6620 0.3380 0.00000 0.00000
## Detection Rate 0.5972 0.1644 0.00000 0.00000
## Detection Prevalence 0.6921 0.2188 0.04398 0.04514
## Balanced Accuracy 0.8106 0.7021 NA NARpart3 - Removing the number of doors variable from the data following the same prediction as above
Variable Importance
## safety persons maint lug_boot buying
## 95.38608 91.93787 57.18082 48.12396 30.21811Confusion matrix
## Confusion Matrix and Statistics
##
## Reference
## Prediction unacc acc good vgood
## unacc 576 15 3 0
## acc 14 164 18 3
## good 0 0 26 7
## vgood 0 8 0 30
##
## Overall Statistics
##
## Accuracy : 0.9213
## 95% CI : (0.9013, 0.9384)
## No Information Rate : 0.6829
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.8349
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: unacc Class: acc Class: good Class: vgood
## Sensitivity 0.9763 0.8770 0.55319 0.75000
## Specificity 0.9343 0.9483 0.99143 0.99029
## Pos Pred Value 0.9697 0.8241 0.78788 0.78947
## Neg Pred Value 0.9481 0.9654 0.97473 0.98789
## Prevalence 0.6829 0.2164 0.05440 0.04630
## Detection Rate 0.6667 0.1898 0.03009 0.03472
## Detection Prevalence 0.6875 0.2303 0.03819 0.04398
## Balanced Accuracy 0.9553 0.9127 0.77231 0.87015Algorithm 2: c50 RULES
Prediction of c50 rules with all the variables
##
## Call:
## C5.0.default(x = train1[, -7], y = train1$Class.Value, rules = TRUE)
##
##
## C5.0 [Release 2.07 GPL Edition] Sun Oct 18 03:05:55 2015
## -------------------------------
##
## Class specified by attribute `outcome'
##
## Read 864 cases (7 attributes) from undefined.data
##
## Rules:
##
## Rule 1: (292, lift 1.4)
## safety = low
## -> class unacc [0.997]
##
## Rule 2: (282, lift 1.4)
## persons = 2
## -> class unacc [0.996]
##
## Rule 3: (54, lift 1.4)
## buying = vhigh
## maint = high
## -> class unacc [0.982]
##
## Rule 4: (18, lift 1.3)
## doors = 2
## persons = more
## lug_boot = small
## -> class unacc [0.950]
##
## Rule 5: (48/2, lift 1.3)
## maint in {vhigh, high}
## lug_boot = small
## safety = med
## -> class unacc [0.940]
##
## Rule 6: (38/6, lift 1.2)
## maint in {vhigh, high}
## doors = 2
## lug_boot = med
## -> class unacc [0.825]
##
## Rule 7: (428/83, lift 1.1)
## buying in {vhigh, high}
## -> class unacc [0.805]
##
## Rule 8: (26, lift 4.4)
## buying in {vhigh, high}
## maint in {med, low}
## persons in {4, more}
## lug_boot = big
## safety in {med, high}
## -> class acc [0.964]
##
## Rule 9: (22, lift 4.4)
## buying = high
## maint in {high, med, low}
## persons in {4, more}
## lug_boot = big
## safety in {med, high}
## -> class acc [0.958]
##
## Rule 10: (45/2, lift 4.3)
## buying in {vhigh, high}
## maint in {med, low}
## persons in {4, more}
## safety = high
## -> class acc [0.936]
##
## Rule 11: (13, lift 4.3)
## buying = low
## maint = high
## persons in {4, more}
## safety = med
## -> class acc [0.933]
##
## Rule 12: (13, lift 4.3)
## buying = med
## maint = high
## persons in {4, more}
## safety = high
## -> class acc [0.933]
##
## Rule 13: (11, lift 4.2)
## buying = high
## maint = high
## persons in {4, more}
## safety = high
## -> class acc [0.923]
##
## Rule 14: (6, lift 4.0)
## buying in {vhigh, high}
## maint in {high, med, low}
## doors = 3
## persons = more
## lug_boot = med
## safety in {med, high}
## -> class acc [0.875]
##
## Rule 15: (4, lift 3.8)
## buying = low
## maint in {vhigh, high}
## doors = 2
## persons in {4, more}
## lug_boot = med
## safety in {med, high}
## -> class acc [0.833]
##
## Rule 16: (9/1, lift 3.8)
## buying in {vhigh, high}
## maint in {high, med, low}
## doors in {4, 5more}
## persons in {4, more}
## lug_boot = med
## safety = med
## -> class acc [0.818]
##
## Rule 17: (194/89, lift 2.5)
## buying in {med, low}
## persons in {4, more}
## safety in {med, high}
## -> class acc [0.541]
##
## Rule 18: (14/1, lift 26.1)
## buying = low
## maint in {med, low}
## persons in {4, more}
## lug_boot in {big, med}
## safety = med
## -> class good [0.875]
##
## Rule 19: (14/2, lift 24.2)
## buying in {med, low}
## maint = low
## persons in {4, more}
## lug_boot in {big, med}
## safety = med
## -> class good [0.813]
##
## Rule 20: (8/1, lift 23.8)
## buying in {med, low}
## maint = low
## persons in {4, more}
## lug_boot = small
## safety = high
## -> class good [0.800]
##
## Rule 21: (8/1, lift 23.8)
## buying = low
## maint in {med, low}
## persons in {4, more}
## lug_boot = small
## safety = high
## -> class good [0.800]
##
## Rule 22: (34/5, lift 20.0)
## buying in {med, low}
## maint in {med, low}
## persons in {4, more}
## lug_boot in {big, med}
## safety = high
## -> class vgood [0.833]
##
## Rule 23: (9/2, lift 17.5)
## buying = low
## maint = high
## persons in {4, more}
## lug_boot in {big, med}
## safety = high
## -> class vgood [0.727]
##
## Default class: unacc
##
##
## Evaluation on training data (864 cases):
##
## Rules
## ----------------
## No Errors
##
## 23 12( 1.4%) <<
##
##
## (a) (b) (c) (d) <-classified as
## ---- ---- ---- ----
## 607 3 1 (a): class unacc
## 1 183 2 2 (b): class acc
## 26 3 (c): class good
## 36 (d): class vgood
##
##
## Attribute usage:
##
## 71.99% buying
## 70.02% safety
## 66.20% persons
## 34.95% maint
## 25.58% lug_boot
## 8.22% doors
##
##
## Time: 0.0 secs## Confusion Matrix and Statistics
##
## Reference
## Prediction unacc acc good vgood
## unacc 596 3 0 0
## acc 3 184 7 2
## good 0 0 34 6
## vgood 0 0 0 29
##
## Overall Statistics
##
## Accuracy : 0.9757
## 95% CI : (0.9631, 0.9849)
## No Information Rate : 0.6933
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9479
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: unacc Class: acc Class: good Class: vgood
## Sensitivity 0.9950 0.9840 0.82927 0.78378
## Specificity 0.9887 0.9823 0.99271 1.00000
## Pos Pred Value 0.9950 0.9388 0.85000 1.00000
## Neg Pred Value 0.9887 0.9955 0.99150 0.99042
## Prevalence 0.6933 0.2164 0.04745 0.04282
## Detection Rate 0.6898 0.2130 0.03935 0.03356
## Detection Prevalence 0.6933 0.2269 0.04630 0.03356
## Balanced Accuracy 0.9918 0.9831 0.91099 0.89189C50 tree with variables- buying and maint price, safety and seating capacity
##
## Call:
## C5.0.default(x = train1[, c(-3, -5, -7)], y = train1$Class.Value)
##
##
## C5.0 [Release 2.07 GPL Edition] Sun Oct 18 03:05:55 2015
## -------------------------------
##
## Class specified by attribute `outcome'
##
## Read 864 cases (5 attributes) from undefined.data
##
## Decision tree:
##
## safety = low: unacc (292)
## safety in {med,high}:
## :...persons = 2: unacc (195)
## persons in {4,more}:
## :...buying in {vhigh,high}:
## :...maint = vhigh: unacc (45)
## : maint in {high,med,low}:
## : :...safety = med: unacc (71/29)
## : safety = high:
## : :...maint in {med,low}: acc (45/2)
## : maint = high:
## : :...buying = vhigh: unacc (11)
## : buying = high: acc (11)
## buying in {med,low}:
## :...safety = high:
## :...maint = vhigh: acc (29/1)
## : maint in {med,low}: vgood (48/19)
## : maint = high:
## : :...buying = med: acc (13)
## : buying = low: vgood (11/4)
## safety = med:
## :...buying = med: acc (46/19)
## buying = low:
## :...maint = high: acc (13)
## maint in {med,low}: good (23/10)
## maint = vhigh:
## :...persons = 4: unacc (7/2)
## persons = more: acc (4)
##
##
## Evaluation on training data (864 cases):
##
## Decision Tree
## ----------------
## Size Errors
##
## 16 86(10.0%) <<
##
##
## (a) (b) (c) (d) <-classified as
## ---- ---- ---- ----
## 590 18 2 1 (a): class unacc
## 31 139 8 10 (b): class acc
## 4 13 12 (c): class good
## 36 (d): class vgood
##
##
## Attribute usage:
##
## 100.00% safety
## 66.20% persons
## 43.63% buying
## 38.31% maint
##
##
## Time: 0.0 secs## Confusion Matrix and Statistics
##
## Reference
## Prediction unacc acc good vgood
## unacc 575 20 0 4
## acc 40 134 12 10
## good 0 9 13 18
## vgood 0 0 0 29
##
## Overall Statistics
##
## Accuracy : 0.8692
## 95% CI : (0.8449, 0.891)
## No Information Rate : 0.7118
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.7157
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: unacc Class: acc Class: good Class: vgood
## Sensitivity 0.9350 0.8221 0.52000 0.47541
## Specificity 0.9036 0.9116 0.96782 1.00000
## Pos Pred Value 0.9599 0.6837 0.32500 1.00000
## Neg Pred Value 0.8491 0.9566 0.98544 0.96168
## Prevalence 0.7118 0.1887 0.02894 0.07060
## Detection Rate 0.6655 0.1551 0.01505 0.03356
## Detection Prevalence 0.6933 0.2269 0.04630 0.03356
## Balanced Accuracy 0.9193 0.8668 0.74391 0.73770SUMMARY
| Algorithm | Variables | Accuracy |
|---|---|---|
| rpart | safety+Persons | 0.7616 |
| rpart | all | 0.941 |
| rpart | all except doors | 0.9213 |
| c50rules | all | 0.9757 |
| c50tree | buying+maint+safety+persons | 0.8692 |
INFERENCE
- All the variables are importance for customers in assessing whether the car is in acceptable or unacceptable range
- Safety and Seating capacity are two main factors in rejecting the cars as unacceptable
- Number of doors are the least important variable in deciding the class value of the car