An Innovative Approach to Predict Bankruptcy

VIVA-Tech International Journal for Research and Innovation Volume 1, Issue 2 (2019)
ISSN(Online): 2581-7280 Article No. 8
PP 1-6
1
www.viva-technology.org/New/IJRI
An Innovative Approach to Predict Bankruptcy
Mihir H. Panchal1
, Mayur B. Bodar1
, Sunny R. Maurya1
, Tatwadarshi P.
Nagarhalli1
1
(Computer Engineering Department, VIVA Institute of Technology, India)
Abstract: Bankruptcy is a legal status of a person or other organization that cannot repay their debts to
creditors. Bankruptcy prediction is the task of predicting bankruptcy and by doing various surveys we can avoid
financial distress of firms. It is a huge area of accounting and finance research. The significance of this area is
an important part of financial specialists and creditors in assessing the probability that a firm may go bankrupt
or not. Estimating the risk of corporate bankruptcies is very important as the effect of bankruptcy is on a global
level. The aim of predicting financial distress is to develop a predictive model that combines various economic
factors which allow foreseeing the financial status of a firm. In this domain, various methods were proposed that
were based on neural networks, Support Vector Machines, Decision Trees, Random Forests, Naïve Bayes,
Balanced Bagging and Logistic Regression. In this paper, we document our observations as we explore and build
a Restricted Boltzmann Machine to Bankruptcy Prediction. We started by carrying out data pre-processing where
we impute the missing data values using Mean Imputation. To solve the data imbalance issue, we apply the
Synthetic Minority Oversampling Technique (SMOTE) to oversample the minority class labels. Finally, we
analyze and evaluate the performance of the model.
Keywords – Artificial Neural Network, Decision Trees, Logistic Regression, Naïve Bayes, Random Forests,
Restricted Boltzmann machine, Support Vector Machine.
1. INTRODUCTION
In recent years, the problem of corporate bankruptcy has attracted the attention of many stakeholders in
the financial sectors such as business investors, market analyst, banking sectors, lawmakers and shareholders.
Predicting bankruptcy is not only important for decision making in financial institutions but also determines the
country’s financial distress to some extent as wrong decision-making in financial institutions can have a
catastrophic effect on national or sometimes a global scale [1].
Predicting bankruptcy is of great importance in financial decision making. No matter how big or small a
company is when it goes bankrupt it affects everyone on a global level and hence prediction of enterprise
bankruptcies is very necessary. Researchers have been working on this domain for a good amount of time. They
tried to understand the reasons for bankruptcy and eventually trying to avoid bankruptcy. The research on this
domain is also functional to the availability of data.
The aim of the bankruptcy prediction is to predict the financial condition of a company and its future
perspectives within the context of long-term operation on the market. It is a vast area of finance and econometrics
that combines knowledge about the historical data of prosperous and unsuccessful companies. Typically,
enterprises are quantified by numerous indicators that describe their business condition that is further used to
induce a mathematical model using past observations.
There are many related works which focus on predicting the best model for the given financial decision-
making problem. The proposed system not simply examine the probability of liquidation but also to analyze the
best training algorithm. The best training model will be a retreat based on the highest classification rate. In the
proposed method system, we use “Restricted Boltzmann machine”.
The dataset is based on the Polish companies’ bankruptcy data analyzed over 2000-2012. The
informational collection is assembled by Emerging Markets Information Services (EMIS).

PP 1-6
2
2. RELATED WORKS
G. Pranav and K. Govinda [1] have proposed to predict bankruptcy using Artificial Neural Network
(ANN). The specialty of this model is that it is able to take into account past experiences and hence make a more
accurate decision over a period of time. Here, they have used Random Forest as a learning algorithm. The mainly
operate by creating decision trees during the training time and outputting the class that is the mode of the classes
(classification) or mean prediction (regression) of the individual trees. The dataset that G. Pranav and K. Govinda
[1] have used is Polish Companies’ analyzed data which has 65 attributes but to predict bankruptcy they have
considered only 3 such as Solvency, Earnings before Interest and Taxes (EBIT) and Liquidity.
Z. Fatima and S. Achchab [2] analyzed suppliers and customers’ payment delays and based on that they
predicted the failure. In this study, Z. Fatima and S. Achchab [2] have used Multivariate Discriminant Analysis,
Logistic Regression, and Decision Trees. The database used contains annual financial statements data for a sample
of Moroccan firms which has a lot of missing data leading to misleading or improper prediction. Z. Fatima and S.
Achchab [2] used different methods and tried to make work for predicting bankruptcy with the limited dataset.
Y. Zaychenko [3] applied fuzzy neural networks adaptive neuro-fuzzy inference system (ANFIS) and
Takagi-Sugeno-Kang (TSK) and fuzzy Group Method of Data Handling (GMDH) is used. The probably of work
under fuzzy and incomplete data to use expert knowledge in a form of fuzzy methods. For training, they used 115
banks of Europe and the testing sample was of 50 banks
S. Fan et. al. [12] have proposed anomaly detection method to detect bankruptcy with Multivariate
Gaussian distribution, One-Class SVM and Isolation Forest. The experiment dataset is from the UCI machine
learning repository. It describes bankrupt about the financial condition of Polish companies. 1st Year, 2nd Year,
3rd Year, 4th Year and 5th Year cases with respect to the ratio of positive to negative samples are Ratio 1:25,
1:24, 1:20, 1:18 and 1:13 respectively.
C. Cheng and C. Chan [5] have proposed financial distress prediction (FDP) has become increasingly
essential in resolving corporate financial risk. Five classification methods utilized to identify financial distress are
Decision tree C4.5, IBK, SVM, Random Forest, and RBF Network. Data pre-processing the financial database
comes from Taiwan Economic Journal (TEJ) Corporation. The proposed system focuses on the feature selection
and variable selection to predict bankruptcy with special optimizing techniques.
Y. Lu, et. al. [6] is to focus on predicting bankruptcy with the new model of SVM augment by SPSO.
The core merits in it make full of use advantage of Switching PSO Algorithm (SPSO) to search for optimal kernel
parameter of SVM. The Data sets that they have used are from the UCI Machine Learning Repository donated on
9th Feb 2014, which consists of 143 Non-Bankruptcy samples and 107 Bankruptcy sample, the total number is
250 sample data sets. Each attribute contains three parameters.
The dataset that M. Wagle et. al. [7] have contains 240 cases, 112 of which are bankrupted cases and 128
are successful cases. Observations come from two to five years before bankruptcy took place. Among these
companies, 56 went bankrupt two to five years later. The data from the first year of all the 120 companies will be
used as training set to train the classifier and the second-year data of all the 120 companies will be used as the
testing set to test the classifier. M. Wagle et. al. [7] have used five classifiers for the attribute selection process
which were Bayesian network, decision tree, logistic regression, neural networks, and SVM. To increase the
accuracy of the prediction models, they found the two common techniques in WEKA called the boosting technique
and the bagging technique, are applied onto the classifiers to increase their prediction accuracy.
G. Kumar et. al. [8] objective of bankruptcy prediction was to determine whether an organization or
financial firm will go bankrupt or not. The objective of boosting algorithm is to assign more weights to the
misclassified instances so that the learner can focus more on them for the succeeding round to classify it
accurately. The dataset they had obtained from 500 French industrial firms during the year 2002 and 2003. The
proposed system focuses on various boosting algorithms like Logit-Boost, NFS-Boost to predict bankruptcy which
gives better results after comparing with the other boosting algorithms.
A. Aghaie et. al. [9] paper proposes that several significant methodological issues were related to the use
of naive Bayes Bayesian Network (BN) models to predict the bankruptcy. The proposed system focuses on two
methods of Bayes Bayesian Network to predict bankruptcy in which first method, only variables that have
important correlations with the variable of interest and second method they investigated the impact on a naive
Bayes model’s performance of the number of states into which continuous variables were discretized.
D. Kang et. al. [10] paper proposes that the genetic algorithm-based coverage optimization techniques of
the SVM ensemble to solve the multicollinearity problem. Considering this background, D. Kang et.al. [10]
focuses on the genetic algorithm-based coverage optimization techniques of the SVM ensemble (CO-SVM) to
predict bankruptcy which gives accurate results than the SVM & DT.
E. Zibanezhad et. al. [11] have used Clementine software and the method of classification and regression
tree are used for mining financial variables. The data collected from financial statements of firms accepted in
Tehran Stock Exchange (TSE) from 1996 to 2009 which contained a total of 25 required financial ratios. The

PP 1-6
3
structure of the decision tree is a tree structure similar to a flowchart the biggest tie in a tree is the root tie and the
leaf ties show groups or group distribution.
3. PROPOSED SYSTEM
The paper proposes Bankruptcy prediction using Restricted Boltzmann Machine (RBM). Restricted
Boltzmann Machines (RBMs) are neural networks that belong to so called Energy Based Models. This type of
neural networks may be not that familiar, yet this kind of neural networks gained big popularity in recent years in
the context of the Netflix Prize where RBMs achieved state of the art performance in collaborative filtering and
have beaten most of the competition. A restricted Boltzmann machine (RBM) is generative stochastic artificial
neural network that can learn a probability distribution over its set of inputs.
Polish dataset has been used for proposed system. It is hosted by the University of California Irvine (UCI)
Machin Learning Repository which is a huge repository of freely accessible datasets for research and learning
purposes for the Machine Learning/Data Science community. This information was collected from the Emerging
Markets Information Service [6] (EMIS), which is a database containing data on developing markets far and wide.
The bankrupt organizations were examined in the period 2000-2012, while the as yet working organizations were
assessed from 2007 to 2013.
Table 1: Summary of the Polish bankruptcy dataset
Dataset characteristic Multivariate
Number of Features 64
Number of Instances
Data
Total Instances Bankrupt
instances
Number of
Instances
1st
year 7027 271 6756
2nd
year 10173 400 9773
3rd
year 10503 495 10008
4th
year 9792 515 9227
5th
year 5910 410 5500
Feature characteristics Real values
Has missing data? Yes
Associated tasks Classification
Date donated 04-11-2016
In table 2, 2nd
column display, the total number of instances in all dataset and 3rd
column display the
number of instances or rows with missing values for at least one of the features 4th
Column display the number of
instances that would remain in each dataset if all rows with missing values were dropped. 5th
Column displays the
percent of data loss if all the rows with missing data values were dropped. As the data loss rate in most of the
datasets is more than 50%, it is now clear that we cannot simply drop the rows with missing values, as it leads to
a loss in the representativeness of data.
Table 2: Missing Data
Data
Set
#Total
Instances
# Instances
with
missing
values
# Instances that would
remain if all rows with
missing values were
dropped
% Data loss if rows
with missing values
were dropped
Year 1 7027 3833 3194 54.54 %
Year 2 10173 6085 4088 59.81 %
Year 3 10503 5618 4885 53.48 %
Year 4 9792 5023 4769 51.29 %
Year 5 5910 2879 3031 48.71 %
There are mainly three issues comes in missing data
1. Missing data information can introduce a major amount of bias.
2. Handling and analysis are of the data more difficult.
3. Create more reductions in efficiency.

PP 1-6
4
Table 3 display below summarizes the populations of class labels. 2nd
Column displays the total instances,
while the 3rd
Column and 4th
Column display the total number of instances with the class as Bankrupt and Non-
Bankrupt respectively. While 5th
Column shows the population percentage of the minority class, i.e., the
Bankruptcy class label, among the total population of the dataset. These numbers in column 5 display that there
is a huge data imbalance.
Table 3: Data Imbalance
Data
Set
# Total
Instances
# Bankrupt
instances in
this forecasting
period
# Non-
Bankrupt
instances in
this forecasting
period
Percentage of
minority class
samples
Year 1 7027 271 6756 3.85 %
Year 2 10173 400 9773 3.93 %
Year 3 10503 495 10008 4.71 %
Year 4 9792 515 9277 5.25 %
Year 5 5910 410 5500 6.93 %
To deal with missing data in the proposed system mean imputation is used. It is the process in which all
the missing value replace with their mean of that context variable. Once completed dataset will not contain any
missing values. Data imbalance will be dealt by oversampling the dataset using Synthetic Minority Oversampling
Technique (SMOTE). It is a majorly used oversampling technique.
Figure 1 shows the System flow of proposed system. In this above-mentioned dataset will be taken and
pre-processing will be done on that to create balanced data. This balanced data will not have any missing data and
will be balanced for both the instances. This balanced data now will be used to train Restricted Boltzmann
Machine.
Figure 1: Proposed System
As the dataset used contains 64 attributes. Users will have to upload provided .csv file for prediction
which contains the 64 attributes with their descriptions.
4. RESULTS AND ANALYSIS
Proposed system outperformed all the previously implemented models with giving much higher accuracy of
70.86%. While previously implemented best model i.e. Support Vector Machine (SVM) struggled to give
accuracy higher than 60%. Hence our proposed model is more efficient to predict bankruptcy and gives much

PP 1-6
5
better performance and result. Table 4 shows the comparison of accuracy of different models and proposed
system. In this we can clearly see that proposed system has higher accuracy than other models. Based on this
predictions user can take required actions and by using proposed system for prediction user can foresee the
financial status of company.
Table 4: Comparative Analysis
Model Dataset Accuracy
Multivariate Gaussian
distribution (MG)
Polish Dataset 0.89
Support Vector Machine
(SVM)
Polish Dataset 0.8975
RBM
(Proposed)
Polish Dataset 0.9614
4. CONCLUSION
In global level bankruptcy is the major issue. Bankruptcy prediction is very important for corporate world
and it must be done to reduce financial distress. Models that have been implemented before were only able to give
results around 60% which is less for such a big problem. The dataset that has been used in proposed system had a
lot of missing data and data that was in that was also unbalanced that is the companies that went bankrupt were
very less as compared to compares to the companies which didn’t go bankrupt. To deal with missing values Mean
imputation technique was used and to deal with unbalanced data issue Synthetic Minority Oversampling
Technique (SMOTE) was used. After pre-processing was done model was trained and it gave 70.86% which is
higher than all the other models that were created before as they were only able to give results around 60%. Most
was given by Support Vector Machine (SVM) at 63.8%. Having a better model and being able to foresee the
financial status of a company or organization can help in avoiding global financial problem.
REFERENCES
[1] G. Pranav, K. Govinda, “Bankruptcy Prediction Using Neural Network”, International Conference on Inventive Systems and Control
(ICISC), 2018, pp. 248-251.
[2] Z. Fatima, S. Achchab, “The impact of payment delays on bankruptcy prediction”, 3rd International Conference of Cloud Computing
Technologies and Applications, 2017.
[3] Y. Zaychenko, “Banks bankruptcy risk forecasting with application of FNN”, 11th International Scientific and Technical Conference
Computer Sciences and Information Technologies, 2016, pp.196-199.
[4] S. Karlos, S. Kotsiantis, N. Fazakis, “Effectiveness of semi-supervised learning in bankruptcy prediction”, 7th International
Conference on Information”, Intelligence, Systems & Applications (IISA), 2016.
[5] C. Cheng, C. Chan, “An attribute selection-based classifier to predict financial distress”, 12th International Conference on Natural
Computation, Fuzzy Systems and Knowledge Discovery, 2016, pp. 1119-1122.
[6] Y. Lu, J. Zhu, N. Zhang, “A hybrid switching PSO algorithm and support vector machines for bankruptcy prediction”, International
Conference on Mechatronics and Control (ICMC), 2014, pp. 1329-1333.
[7] M. Wagle, Z. Yang, Y. Benslimane, “Bankruptcy prediction using data mining techniques”, 8th International Conference of
Information and Communication Technology for Embedded Systems (IC-ICTES), 2017.
[8] G. Kumar, S. Roy, “Development of Hybrid Boosting Technique for Bankruptcy Prediction”, International Conference on Information
Technology (ICIT), 2016, pp. 248-P253.
[9] A. Aghaie, A. Saeedi, “Using Bayesian Networks for Bankruptcy Prediction: Empirical Evidence from Iranian Companies”,
International Conference on Information Management and Engineering, 2009, pp. 450-455.
[10] D. Kang, M. Kim, “Performance enhancement of SVM ensembles using genetic algorithm in bankruptcy prediction”, 3rd International
Conference on Advanced Computer Theory Engineering and (ICACTE), 2010, pp- V2-154 – V2-158.

PP 1-6
6
[11] E. zibanezhad, D. Foroghi, A. Monadjemi, “Applying decision tree to predict bankruptcy”, IEEE International Conference on
Computer Science and Automation Engineering, 2011, pp. 165-169.
[12] S. Fan, G. Liu, Z. Chen, “Anomaly detection method for bankruptcy prediction”, 4th International Conference on Systems and
`Informatics (ICSAI), 2011, pp. 1456-1460.
[13] https://en.wikipedia.org/wiki/Machine_learning , Last Accessed on 15th
Sept. 2018.
[14] https://en.wikipedia.org/wiki/Linear_regression , Last Accessed on 16th
Sept. 2018.
[15] https://en.wikipedia.org/wiki/Random_forest , Last Accessed on 16th
Sept. 2018.

An Innovative Approach to Predict Bankruptcy

Related slideshows

Recommended for you

More Related Content

What's hot

What's hot (20)

Similar to An Innovative Approach to Predict Bankruptcy

Similar to An Innovative Approach to Predict Bankruptcy (20)

More from vivatechijri

More from vivatechijri (20)

Recently uploaded

Recently uploaded (20)

An Innovative Approach to Predict Bankruptcy