In Partial Fulfillment of the Requirements for the Course

CE505 – CONSTRUCTION METHODS AND PROJECT MANAGEMENT

Published Journal Article Critical Review Sheet

TITLE OF ARTICLE Predicting business failure of construction contractors using long short-term memory
recurrent neural network(LSTM RNN)

AUTHOR/S Youjin Jang , In-Bae Jeong , Yong K. Cho and Yonghan Ahn

JOURNAL PUBLISHER Journal of Construction Engineering and Management

YEAR PUBLISHED 2019

ARTICLE’S PROBLEM STATEMENT

Business Failure Prediction has been one of the most common problems of contractors. Since contractors are
vulnerable to failures in the construction industry due to some factors like the type of project, project schedule,
complexity of project and the risk in implementing construction activities. Construction contractors are increasingly
likely to fail in their business because of the significant structural changes such as globalization, increased competition
and regulation.

ARTICLE’S MAIN GOAL / OBJECTIVE / PURPOSE

Since contractors are more likely fail nowadays due to changes in the technological and global evolution this study
aims to develop new solutions that can help in predicting business failures and help them address the problems before
it ruins the project. It aims to avoid high probability of failures encountered by contractors.

ARTICLE’S SPECIFIC GOAL OBJECTIVE/S

The study aims to develop hybrid methodologies as an effort to improve prediction performance. Also another goal of
this article is to predict business failure within one year with the use of  Recurrent neural network, which is a neural
network with recurrent connections, has been developed in order to learn sequential and temporal patterns from time-
series.

ARTICLE’S SIGNIFICANCE
This study considers that the characteristics of the construction industry are reflected in the value of the financial ratio
of the construction contractors, and twelve financial ratios are selected as input variables : return on asset; return on
equity; return on sales; current ratio; current assets to net assets; working capital to total asset; total liabilities to net
worth; retained earnings to sales; debt ratio; working capital turnover; equity turnover; and total asset turnover.

ARTICLE’S TYPE OF INQUIRY ( PUT  TO THE TYPE OF RESEARCH)

 Quantitative research
 Qualitative research 
 Quantitative and qualitative research or Mixed Method research

ARTICLE’S RESEARCH DESIGN

The current study utilized performance of the prediction model using long short-term and memory recurrent neural
network, Feedforward Neural Network were used as a comparison reference in obtaining data for the research.

2
ARTICLE’S SPECIFIC DATA COLLECTED
The current study gathered data from the respondent about their healthy observations and business failure
observations when considering all the firm-year data. This study used a five-fold cross-validation approach when
examining the prediction performance.

ARTICLE’S METHOD OF DATA ANALYSIS

This study used SMOTE + Tomek links, which is a hybrid method that combines the SMOTE (Synthetic Minority Over-
sampling Technique) and Tomek links to obtain a balanced dataset. SMOTE is an over-sampling method where the
minority class is oversampled to generate new synthetic examples. To measure the performance of the classification
techniques the accuracy and F1-score are used.

SUMMARY OF THE ARTICLE’S FINDINGS

The business failure of construction contractors may cause not only sizeable losses to the economy but also another
business failure to various stakeholders. Therefore, forecasting the potential business failure of construction
contractors has always been an valuable issue for both researchers and practitioners. To make a robust model to
enable the business failure prediction of construction contractors, this study applied long short-term and memory
recurrent neural network which has a time step of three. LSTM RNN feeds the output of the previous time step to the
input of the next step. LSTM RNN has achieved great success in various applications on sequential data because of
the gate and memory mechanism. These results show that the LSTM RNN model is a reliable technique to forecast the
business failure of construction contractors.

CONTRIBUTION OF THE ARTICLE’S FINDINGS TO THE CURRENT LITERATURE

The current study’s findings can be used to improve the understanding of future readers regarding the use of long
short-term and memory recurrent neural network.  The proposed LSTM RNN model is a alternative to assist managers,
investors, auditors, and government entities in predicting business failure of construction contractors which can be
also adapted to other industry cases.

IMPLICATION OF THE ARTICLE’S FINDINGS TO CURRENT PRACTICE

This study proposed the business failure prediction model of construction contractors using LSTM RNN. Twelve
financial ratios of the listed U.S. construction contractors were used to develop a model which can predict business
failures within one year.

AREA/S FOR FUTURE RESEARCH SUGGESTED BY THE ARTICLE

There is still room for improvement for predicting performance and applicability of the LSTM RNN model for business
failure prediction. The LSTM RNN model needs to cover a relatively long project period in the construction industry
including managerial and economic variables. It also remains an interesting topic for further study to derive
optimization producers to select proper group of variables and the corresponding parameter values.

NAME OF STUDENT REYNALDO BONIFACIO GERMAR

DATE SUBMITTED MARCH 19,2020
LEVEL OF ASSESSMENT BEST BETTER GOOD POOR
(PUT A  TO THE EXTENT OF 
YOUR REVIEWSHIP)
Percentage equivalent 99-100 90-98 81-89 75-80
GPA equivalent 1.00 1.25-1.75 2.00-2.50 2.75-3.00

2
2