This analysis aims to accurately predict whether a business will declare bankruptcy or not. The dataset is available here: https://www.kaggle.com/shebrahimi/financial-distress/data
library(caret)
library(mlbench)
library(DMwR)
library(e1071)
library(randomForest)
library(dplyr)
library(plotly)
df <- read.csv("Financial Distress.csv", stringsAsFactors=FALSE)
#check the dimensions of the dataset
dim(df)## [1] 3672 86#take a look at column data types
str(df)## 'data.frame': 3672 obs. of 86 variables:
## $ Company : int 1 1 1 1 2 2 2 2 2 2 ...
## $ Time : int 1 2 3 4 1 2 3 4 5 6 ...
## $ Financial.Distress: num 0.0106 -0.456 -0.3254 -0.5666 1.3573 ...
## $ x1 : num 1.28 1.27 1.05 1.11 1.06 ...
## $ x2 : num 0.02293 0.00645 -0.05938 -0.01523 0.10702 ...
## $ x3 : num 0.875 0.821 0.922 0.859 0.815 ...
## $ x4 : num 1.216 1.005 0.729 0.81 0.836 ...
## $ x5 : num 0.0609 -0.0141 0.0205 0.076 0.2 ...
## $ x6 : num 0.1883 0.181 0.0449 0.091 0.0478 ...
## $ x7 : num 0.525 0.623 0.433 0.675 0.742 ...
## $ x8 : num 0.01885 0.00642 -0.08142 -0.01881 0.12803 ...
## $ x9 : num 0.183 0.036 -0.765 -0.108 0.577 ...
## $ x10 : num 0.00645 0.0018 -0.05432 -0.06532 0.09408 ...
## $ x11 : num 0.858 0.852 0.893 0.896 0.815 ...
## $ x12 : num 2.006 -0.486 0.412 0.995 3.015 ...
## $ x13 : num 0.1255 0.1793 0.0776 0.1411 0.1854 ...
## $ x14 : num 6.97 4.58 11.89 6.09 4.39 ...
## $ x15 : num 4.65 3.75 2.49 1.64 1.62 ...
## $ x16 : num 0.0501 -0.014 0.0281 0.0939 0.2392 ...
## $ x17 : num 2.2 2.46 1.4 2.06 3.03 ...
## $ x18 : num 0.01826 0.02756 0.0126 0.0116 0.00681 ...
## $ x19 : num 0.025 0.0288 0.0681 0.0944 0.0793 ...
## $ x20 : num 0.02726 0.0411 0.01485 0.01442 0.00888 ...
## $ x21 : num 1.417 1.18 0.817 0.904 1.025 ...
## $ x22 : num 9.56 7.3 7.12 7.98 4.75 ...
## $ x23 : num 0.1487 0.056 0.0652 0.1252 0.266 ...
## $ x24 : num 0.67 0.67 0.848 0.805 0.768 ...
## $ x25 : num 214.8 38.2 -498.4 -75.9 1423.1 ...
## $ x26 : num 12.6 12.9 13.2 13.3 11.6 ...
## $ x27 : num 6.46 5.55 16.25 8.89 17.49 ...
## $ x28 : num 0.0438 0.2655 0.4166 0.0838 0.6208 ...
## $ x29 : num 0.2046 0.1502 0.0741 0.0541 0.0469 ...
## $ x30 : num 0.352 0.418 0.367 0.544 0.57 ...
## $ x31 : num 8.32 9.53 9.35 7.09 9.49 ...
## $ x32 : num 0.289 0.416 0.504 0.671 0.681 ...
## $ x33 : num 0.766 0.817 0.92 0.937 0.942 ...
## $ x34 : num 2.58 2.6 1.49 2.35 4.13 ...
## $ x35 : num 77.4 95.9 144.7 219.8 222.7 ...
## $ x36 : num 0.02672 0.00758 -0.06648 -0.017 0.13123 ...
## $ x37 : num 1.631 0.838 0.956 0.383 0.253 ...
## $ x38 : num 0.01502 0.02743 0.01727 0.01433 0.00815 ...
## $ x39 : num 0.00548 0.04543 0.02806 0.20337 0.35301 ...
## $ x40 : num 0.127 0.138 0.102 0.101 0.176 ...
## $ x41 : num 9.7 5.6 9.4 5.74 4.51 ...
## $ x42 : num -0.736 -0.644 -14.032 0.722 -0.113 ...
## $ x43 : num 0.986 1.302 0.757 1.391 1.053 ...
## $ x44 : num 0.1802 0.0469 -0.5798 -0.1501 0.6077 ...
## $ x45 : num 1.501 1.01 0.578 0.645 0.258 ...
## $ x46 : num 0.02622 0.00786 -0.06437 -0.01773 0.13138 ...
## $ x47 : num 7.05 4.6 11.99 6.11 4.42 ...
## $ x48 : num 1175 1062 651 703 2465 ...
## $ x49 : num 5.34 3.74 10.93 5.7 4.14 ...
## $ x50 : num 0.851 0.944 0.935 0.875 0.734 ...
## $ x51 : num 12.8 12.9 12.9 13.1 11.4 ...
## $ x52 : num 0.061737 -0.000565 0.041625 0.1084 0.25031 ...
## $ x53 : num 0.1809 0.0563 0.0476 0.1013 0.2224 ...
## $ x54 : num 210 250 281 414 315 ...
## $ x55 : num -0.5826 -0.4748 -1 0.565 -0.0601 ...
## $ x56 : num 0.471 0.386 0.488 0.344 0.202 ...
## $ x57 : num 0.1099 0.3693 0.0533 0.0734 1.2291 ...
## $ x58 : num 0.00 0.00 3.79e-03 3.66e-05 -2.49e-03 ...
## $ x59 : num 0.00 0.00 5.19e-03 4.53e-05 -2.98e-03 ...
## $ x60 : num 0.22 0 0 0 0.227 ...
## $ x61 : num 7.12 7.42 3.64 5.14 7.12 ...
## $ x62 : num 15.38 7.11 7.02 9.91 15.38 ...
## $ x63 : num 3.27 14.32 1.15 2.04 3.27 ...
## $ x64 : num 17.87 18.77 9.9 -1.49 17.87 ...
## $ x65 : num 34.69 124.76 6.45 -21.91 34.69 ...
## $ x66 : num 30.1 26.1 30.2 34.3 30.1 ...
## $ x67 : num 12.8 11.8 10.3 11.5 12.8 11.8 10.3 11.5 11.3 10.5 ...
## $ x68 : num 7991 8323 8747 9042 7991 ...
## $ x69 : num 364.95 0.19 11.95 -18.75 364.95 ...
## $ x70 : num 15.8 15.6 15.2 10.4 15.8 15.6 15.2 10.4 11.9 18.4 ...
## $ x71 : num 61.5 24.6 20.7 47.4 61.5 ...
## $ x72 : num 4 0 0 4 4 0 0 4 4 2 ...
## $ x73 : num 36 36 35 33 36 36 35 33 31 29 ...
## $ x74 : num 85.4 107.1 120.9 54.8 85.4 ...
## $ x75 : num 27.1 31.3 36.1 39.8 27.1 ...
## $ x76 : num 26.1 30.2 35.3 38.4 26.1 ...
## $ x77 : num 16 17 17 17.2 16 ...
## $ x78 : num 16 16 15 16 16 16 15 16 14 12 ...
## $ x79 : num 0.2 0.4 -0.2 5.6 0.2 0.4 -0.2 5.6 2.1 -6.4 ...
## $ x80 : int 22 22 22 22 29 29 29 29 29 29 ...
## $ x81 : num 0.0604 0.0106 -0.456 -0.3254 1.251 ...
## $ x82 : int 30 31 32 33 7 8 9 10 11 12 ...
## $ x83 : int 49 50 51 52 27 28 29 30 31 32 ...#check for NA values
table(is.na(df))##
## FALSE
## 315792#if target variable "Financial.Distress" is greater than -0.50 the company is healthy (0), otherwise, it is financially distressed (1).
df$Financial.Distress <- ifelse(df$Financial.Distress > -0.5, 0, 1)
df$Financial.Distress <- as.factor(df$Financial.Distress)Since this is a classification problem, I will visualize the class balance with an interactive plotly donut chart:
percs <- df %>%
group_by(Financial.Distress) %>%
summarise(n = n()) %>%
mutate(perc = n/nrow(df) * 100) %>%
mutate(perc = round(perc, digits = 1))
percs$health <- ifelse(percs$Financial.Distress == 0, "Healthy", "Distressed")
percs %>% plot_ly(labels = ~health, values = ~perc) %>%
add_pie(hole = 0.6) %>%
layout(title = "Class balance in target variable", showlegend = T,
xaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE),
yaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE))n <- nrow(df)
trainIndex <- sample(1:n, size = round(0.7*n), replace=FALSE)
train <- df[trainIndex, 3:82]
test <- df[-trainIndex, 3:82]
set.seed(7)
trainControl <- trainControl(method="cv", number=5)
fit.rf <- train(Financial.Distress ~ ., data = train, method="rf", trControl=trainControl)
p <- predict(fit.rf, test)
confusionMatrix(p, test$Financial.Distress)## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 1060 22
## 1 9 11
##
## Accuracy : 0.9719
## 95% CI : (0.9603, 0.9808)
## No Information Rate : 0.9701
## P-Value [Acc > NIR] : 0.40552
##
## Kappa : 0.4016
## Mcnemar's Test P-Value : 0.03114
##
## Sensitivity : 0.9916
## Specificity : 0.3333
## Pos Pred Value : 0.9797
## Neg Pred Value : 0.5500
## Prevalence : 0.9701
## Detection Rate : 0.9619
## Detection Prevalence : 0.9819
## Balanced Accuracy : 0.6625
##
## 'Positive' Class : 0
## At 96.82 percent, the accuracy is deceivingly high, with a very high true negative rate, but a very low true positive rate, as shown by the confusion matrix.
Now, I will oversample the minority class by 200 percent and undersample the majority class by 100 percent in the training set using the SMOTE function in the DMwR package:
n <- nrow(df)
trainIndex <- sample(1:n, size = round(0.7*n), replace=FALSE)
train <- df[trainIndex, 3:82]
test <- df[-trainIndex, 3:82]
trainSplit <- SMOTE(Financial.Distress ~ ., train, perc.over = 200, perc.under = 100)
set.seed(7)
trainControl <- trainControl(method="cv", number=5)
fit.rf <- train(Financial.Distress ~ ., data = trainSplit, method="rf", trControl=trainControl)
p <- predict(fit.rf, test)
confusionMatrix(p, test$Financial.Distress)## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1
## 0 914 7
## 1 143 38
##
## Accuracy : 0.8639
## 95% CI : (0.8422, 0.8836)
## No Information Rate : 0.9592
## P-Value [Acc > NIR] : 1
##
## Kappa : 0.2898
## Mcnemar's Test P-Value : <2e-16
##
## Sensitivity : 0.8647
## Specificity : 0.8444
## Pos Pred Value : 0.9924
## Neg Pred Value : 0.2099
## Prevalence : 0.9592
## Detection Rate : 0.8294
## Detection Prevalence : 0.8358
## Balanced Accuracy : 0.8546
##
## 'Positive' Class : 0
## Here we can see that although the overall accuracy has dropped slightly, the true positive rate in the confusion matrix has risen significantly to accurately predict nearly all positive cases.