Slide Presentation Predicting Airline Passenger Satisfaction using Random Forest Algorithm

Outline

  • What is Random Forest Algorithm?

  • Benefits of Random Forest Algorithm

  • Predicting Airline Passenger Satisfaction using

  • Random Forest Algorithm

List of Packages used

library(readr)
library(DiagrammeR)
library(tidyverse)
library(report)
library(gtsummary)
#library(caret)
library(e1071)
library(sjPlot)
library(performance)
#library(ggstatsplot)
library(SmartEDA)
library(dlookr)
library(lme4)
library(lmerTest)
#library(neuralnet)
#library(DataExplorer)
#library(rpivotTable)
# library(ConfusionTableR)
# library(reshape)
# library(mlbench)
# library(Boruta)
# library(rpart)
# library(rpart.plot)
# library(randomForest)
#library(flextable)
library(readr)
Capd <- read_csv("Capd.csv")

Material and Methodology

  • A secondary data-set from the Kaggle data science platform

  • Machine algorithms Random Forest classifier

  • Data sets were partitioned into training and testing,

  • Of which 70% of the data was retained as a training set while 30% was considered as a testing set.

What is Random Forest Algorithm?

What is Random Forest Algorithm?

What is Random Forest Algorithm?

  • Random Forest Algorithm is an ensemble learning method for classification and regression

  • Combines multiple decision trees to create a more accurate and stable prediction

  • Uses bagging and feature randomness when building each individual tree

Benefits of Random Forest Algorithm

  • High accuracy and robustness
  • Handles missing values and outliers
  • Reduces variance compared to a single decision tree

Predicting Airline Passenger Satisfaction

  • Collect data on airline passengers

  • Use the Random Forest Algorithm to build a model

  • Use the model to predict passenger satisfaction

Reading the Data File

library(readr)
Capd <- read_csv("Capd.csv")

Pre-processing

# Capd %>% na.omit() # Removing missing values in the dataset 
# 
# # Converting data types using the following methods below
# Capd$Gender <- as.factor(Capd$Gender)
# Capd$customer_type <-  as.factor(Capd$customer_type)
# Capd$type_of_travel <- as.factor(Capd$type_of_travel)
# Capd$customer_class <- as.factor(Capd$customer_class)
# Capd$flight_distance <- as.integer (Capd$flight_distance)
# Capd$inflight_wifi_service <- as.factor(Capd$inflight_wifi_service)
# Capd$ease_of_online_booking <- as.factor(Capd$ease_of_online_booking)

Demo-graphical Characteristics

Characteristic N = 129,8801
Gender
    Female 65,899 (51%)
    Male 63,981 (49%)
customer_type
    disloyal Customer 23,780 (18%)
    Loyal Customer 106,100 (82%)
customer_class
    Business 62,160 (48%)
    Eco 58,309 (45%)
    Eco Plus 9,411 (7.2%)
1 n (%)

Estimated Parameters

Capd %>% 
  select(age,flight_distance,cleanliness,departure_delay_in_minutes,arrival_delay_in_minutes) %>% 
na.omit() %>%  report_parameters() 
  - age: n = 129487, Mean = 39.43, SD = 15.12, Median = 40.00, MAD = 17.79, range: [7, 85], Skewness = -3.38e-03, Kurtosis = -0.72, 0% missing
  - flight_distance: n = 129487, Mean = 1190.21, SD = 997.56, Median = 844.00, MAD = 767.99, range: [31, 4983], Skewness = 1.11, Kurtosis = 0.27, 0% missing
  - cleanliness: n = 129487, Mean = 3.29, SD = 1.31, Median = 3.00, MAD = 1.48, range: [0, 5], Skewness = -0.30, Kurtosis = -1.01, 0% missing
  - departure_delay_in_minutes: n = 129487, Mean = 14.64, SD = 37.93, Median = 0.00, MAD = 0.00, range: [0, 1592], Skewness = 6.85, Kurtosis = 101.88, 0% missing
  - arrival_delay_in_minutes: n = 129487, Mean = 15.09, SD = 38.47, Median = 0.00, MAD = 0.00, range: [0, 1584], Skewness = 6.67, Kurtosis = 95.12, 0% missing

Cross-Tabulation of Customers Type

tab_xtab(var.row = Capd$Gender, 
         var.col = Capd$customer_type,
         show.row.prc = T)
Gender customer_type Total
disloyal Customer Loyal Customer
Female 12843
19.5 %		 53056
80.5 %		 65899
100 %
Male 10937
17.1 %		 53044
82.9 %		 63981
100 %
Total 23780
18.3 %		 106100
81.7 %		 129880
100 %
χ2=124.313 · df=1 · φ=0.031 · p=0.000

Cross-Tabulation based on Class

tab_xtab(var.row = Capd$Gender, 
         var.col = Capd$customer_class,
         show.row.prc = T)
Gender customer_class Total
Business Eco Eco Plus
Female 31263
47.4 %		 29670
45 %		 4966
7.5 %		 65899
100 %
Male 30897
48.3 %		 28639
44.8 %		 4445
6.9 %		 63981
100 %
Total 62160
47.9 %		 58309
44.9 %		 9411
7.2 %		 129880
100 %
χ2=20.908 · df=2 · Cramer's V=0.013 · p=0.000

Conceptual Model used

Feature Selection

#FS <- Boruta(satisfaction~., data = Capd, doTrace =2)

Data Partition

# set.seed(1234) # A Random Sampling with replacement
# 
# #Data Partition in to trainign and testing
# 
# Model <- sample(2, nrow(Capd), replace = T, prob = c(0.7, 0.3))
# train <- Capd[Model ==1,]
# test <- Capd[Model ==2,]

Building the Model

# #Random Forest Model
# set.seed(333)
# as.data.frame(Capd) # we converted the data in to datafarme
# 
# rf23 <-randomForest(satisfaction~., data = train, method = "class", na.action=na.exclude)

Training set of the Model

# # Prediction & Confusion Matrix - Test
# p <- predict(rf23, train)
# confusionMatrix(p, train$satisfaction)

Training set of the Model

Testing Set of the Model

# p2 <- predict(rf23, test)
# confusionMatrix(p2, test$satisfaction)

Testing Set of the Model

Confusion Metrix for Training set

# ConfusionTableR::binary_visualiseR(train_labels = train$satisfaction,
#                                    truth_labels= train$satisfaction,
#                                    class_label1 = "Not satisfied",
#                                    class_label2 = "Satisfied",
#                                    quadrant_col1 = "#28ACB4",
#                                    quadrant_col2 = "#4397D2",
#                                    custom_title = "Confusion Metric on Airline P",
#                                    text_col= "black")

Confusion Metrix for Training set

Confusion Metrix for Testing set

# ConfusionTableR::binary_visualiseR(train_labels = test$satisfaction,
#                                    truth_labels= test$satisfaction,
#                                    class_label1 = "Not satisfied",
#                                    class_label2 = "Satisfied",
#                                    quadrant_col1 = "#28ACB4",
#                                    quadrant_col2 = "#4397D2",
#                                    custom_title = "Confusion Metric on Airline P",
#                                    text_col= "black")

Confusion Metrix for Testing set

Conclusion

  • The overall accuracy of the model was found to be 95% with a sensitivity of 97% and specificity of 93%

  • Most of the customers were unsatisfied with the airline service. Therefore based on these findings, the Random-Forest algorism predicated 57%  at 95%  accuracy with a sensitivity of 97% and specificity of 93% that the participants were not satisfied with the daily operation of the airline industry, especially in the areas involved in Air travelers purchasing ticket/booking online, values added services

Recommendation

  • Considering that 57% of participants reported not being satisfied with airline service rendered this is a significant proportion that may significantly decrease the daily, weekly or monthly income revenues generated.

  • Therefore, this findings recommends that the airline industry should endeavor to improve daily operation services, especially in the areas of travelers purchasing tickets/booking online, value-added services such, as In-flight Wi-Fi service, check-in service, Baggage handling

  • This will increase the volume of patronage and, as a result, boost their market share and hence profitability.