Managing of Subcontractors Failure in Construction Projects

A building entity that enters into a contract with a major contractor in order to do certain obligations in a project within the scope of the whole contract, is usually termed as subcontractor. It may provide labourers, deliver equipment and materials, provide tools, and design services for the project [1, 2]. Subcontractors are divided by Mbachu [3] into three groups: 1) Those who perform certain trades like painting or bricklaying are trade contractors; 2) Specialist subcontractors, who perform specialized services like cladding, piping, tailing etc.; and 3) Those who only perform services of labour like foremen and common worker are the labour-only subcontractors. Subcontracting was not used as frequently as it is now in previous decades [4].

Construction of buildings is well known for being a hub of traders with a range of skills. The size and complexity of the projects undertaken, as well as the “one-time off project” nature of its working environment, make the building business unique. Due to the sector’s dynamic character, which needs the acquisition of fresh technologies and cutting-edge procedures in order to enable the supply of goods, the use of subcontractors has become vital [5]. The organization or person in the construction industry who enters into a written agreement with the principal contractor to carry out specific project-related tasks in accordance with the general contract. Beside who may also provide personnel, supplies, machinery, tools, and designs. Laryea [6] revealed that subcontractors sign contracts to carry out a contract’s components or sections.

In order to prevent a domino effect of business failures, the construction sector must overcome significant challenges in predicting firm failure and providing warning signs. The majority of pertinent research have created models that, using financial ratios, forecast the likelihood that a firm would fail within a year. Several research has tried to include non-financial information However, they could not provide any evidence to suggest that this addition may enhance a model’s ability to predict outcomes [7, 8].

Comparatively speaking, the proportion of subcontracted projects has grown recently. Subcontracting can really contribute between 50 and 90 percent, and it may even account for up to 90 percent of the entire project value, during a building process [9, 4]. Therefore, the subcontracting issues are a topic of discussion among the influential people in the construction business. Subcontracting-related problems, on the other hand, are among the construction industry’s key risks that are frequently brought up in international discussions. The date of completing delays [10], the safety of on-site personnel is one of the dangers encountered that is frequently studied [11, 12], and the inability of parties to communicate throughout the phase of construction [13, 14].

It was advised that enterprises employ other variables, such as pricing, quality, delivery time, risk, prior experiences with subcontractors, and reputation, to address the issue of subcontractor selection [15, 16].

The success or failure of the project will depend on the subcontractors’ performance, who handle 85% of the building work [5]. Project completion delays are caused by subcontractors’ poor planning, their incapacity to execute quality assurance and control programs, and job safety laws [17]. According to White and Marasini [18], delays in project completion are brought on by poor communication between the contractor and subcontractors during the building process as well as by work not moving forward according to the original timetable. Regardless of the complexity, scale, or environment in which it is built, a project is deemed successful provided it is finished on schedule, under budget, and with acceptable quality [19].

Effective collaboration enhances project performance by raising profit margins and lowering claims in addition to ensuring timely, affordable, and high-quality completion [20]. Tayeh [21] talked about what causes problems and how long it takes to finish projects in the Gaza Strip. He discovered that certain jobs were transferred to a different subcontractor without notifying the initial subcontractor. Another element influencing a project’s success or failure is the main contractor’s authoritarian attitude toward subcontractors [22]. Effective collaboration and mutual trust are two of the finest markers of the advancement of building projects, according to previous studies [23].

According to Daoud et al. [24], the top five factors contributing to delays in construction projects were, in order of significance, low labor productivity, ineffective coordination and communication between stakeholders, various bribery schemes, project financing delays, work order changes, and unskilled labor. Out of the five types of risks, Adinyira et al. [25] said that financial, resource, technical, and management risks had a beneficial impact on the quality performance of building projects. On a construction project, improper risk management of subcontractors’ risks will have a significant negative impact on the project’s performance quality. According to Elhusseiny [26], low worker productivity, owner payment delays, and delayed or nonexistent delivery of equipment and building supplies all contribute to project delays and interpersonal issues amongst project participants.

In the Pacific Northwest region of the United States, McCord and Gunderson’s study [27] interviews with twenty-four subcontractors revealed that the most serious betrayal of trust, which was enough to end a relationship for 75% of the subcontractors surveyed, was post-award bid shopping by principal contractors. Trust was identified as the fundamental building block of good subcontractor-principal contractor relationships.

The main difficulties that have not been addressed among the enormous amount of research on subcontracting were the subcontractors’ predictions before the specific project is awarded and the standards used to observe the subcontractors throughout the building phase. Consequently, in order to address that, these crucial difficulties will be resolved in this study. The determining of the relationships between the key success variables for subcontractors in the construction sector, as growing engagement of subcontractors underlined the importance of resolving these challenges. In broad terms, an overall contractor can anticipate the advantages and disadvantages of choosing a subcontractor. Once the subcontractor’s performance or weaknesses have been predicted, resource allocation may then be pre-planned. For instance, if a subcontractor’s performance is subpar, additional time, money, and labour resources are required to stabilize the performance. Additionally, principal contractors possess the choice and authority to choose the wanted subcontractors they believe will be able to meet their needs. For instance, subcontractors projected to perform well on “communication skills” throughout the building stage are favoured above others in a project with a high likelihood of redesigning. In conclusion, it is more relevant to forecast performance and manage subcontractors rather than examine subcontractor practice in many aspects separately. The old method of collecting cost information, which is prone to errors, continues to be a challenge for contractor-led projects. This approach has often resulted in the defeat of project objectives, destruction of feasibility investigations that guided project investment, and general undermining of project success [28]. However, with model element features like size, area, object family type, and productivity estimates, digital construction information flow inside the 5D BIM process enables contractors, employers, and project teams to provide precise cost and critical estimation information [29]. Adopting BIM to improve project performance on a single project is challenging. It is considerably more difficult to broadly use methodologies for projecting the influence of BIM on future projects. In order to offer recommendations, some previous studies have employed surveys to ascertain and evaluate project participants’ perceptions of the advantages and return on investment that may be realized by utilizing BIM [30].

The results of the using of each part of the methodology as follow:

3.1 BIM for subcontractor performance

Three case studies were taken for different type of school, 12, 18 and 24 classes, these schools were awarded to subcontractor in order to be implemented and during the progress of work several problems arise, and change orders cost deviation and other. Each case study was modelled both structure and architectural to estimate the different.

Only a few metrics were employed to characterize project performance (PP), including construction management, on-time completion, quality, and cost [31], which are usually referred to as “the iron triangle” by which the construction project’s performance could be evaluated as well as its success degree [18].

Due to the fact that both “schedule” and “cost” could be examined objectively, both elements were designated as the most typical performance indices [32, 33]. Even if the concept of “quality” was difficult to quantify, the pleasure of the project’s owner still depends on this specific building quality. Due to the fact that project managers are accountable for balancing the project’s quality, cost and schedule. Several projects were taken in order to prove the hypothesis which indicate that the subcontractor must use BIM as too for cost estimation. In Figure 2, the building was modelled by using BIM. As Figure 2(a) shows the architectural form of the building and Figure 2(b) the structural form of the building.

21.jpg

(a)

22.jpg

(b)

Figure 2. Structural and architectural model for 12 class

The same was repeated for different projects in Figure 3. Determining the criteria used for observing subcontractors throughout the building phase and evaluating subcontractors’ performance prior to project awarding are hence the research aims of this study.

31.jpg

(a)

32.jpg

(b)

Figure 3. Structural and architectural model for 24 class

As the subcontractor, a several deviations occurs between the tenders and actual work and thus return to the disability of the subcontractor to work as planned. Figure 4 shows the calculation in three different methods, tender which mean the cost as provided originally by the main contractor, estimated which mean the cost calculated by the subcontractor and actual represent the cost calculated by BIM.

41.jpg

(a)

42.jpg

(b)

Figure 4. Structural and architectural model for 18 class

Figures 5 and 6 represent the deviation between the three calculation in curves. After these case studies were taken, it should be noted several issues with subcontractor management and hence need to be predicted to take a preventive action.

5.jpg

Figure 5. Different in quantities for school with 12 class

6.jpg

Figure 6. Different in quantities for school with 18 class

3.2 AI for subcontractor management

The data used for the prediction collected from 234 projects of different types (residential, road, electrical, water projects). Each projects contain the following information (type of project, additional time, additional cost, percentage of completion, change orders, deviation, performance of subcontractors), the factors that used in the prediction as shown in Figure 7.

The algorithms that used for prediction (logistic regression, decision tree and Naive Bayes).

Random forests, sometimes referred to as random selection forests, are a method of ensemble learning of regression, classification, as well as other missions that employs a large number of selection bushes to train before producing the elegance, which is the mean/common prediction (regression) of the individual bushes. The problem of selection bushes overfitting their educational set is solved by random selection forests. In most situations, random forests perform better than preference bushes, although they are less specialized than gradient enhanced bushes. On the other hand, data attributes may also have an impact on how well they work.2020 and 2021, respectively, by Piryonesi. The BCR prediction tool was created in Python and offers prediction models for 4 different sets of rules [34, 35].

During the training phase, every feature variable in the data was employed. The predicted events are listed, and the confusion matrix’s findings are made public. As can be observed, the great majority of estimated transactions are contained in the “True-True” category. Regression using logit. A sigmoid function that depicts it is illustrated by Figure 8.

7.jpg

Figure 7. Prediction variables

8.jpg

Figure 8. Different in quantities for school with 24 class

9.jpg

Figure 9. Logistic regression [36]

To anticipate the output values, weights or coefficients are blended linearly with the input values. It designates a binary variable with simply the 0 and 1 possible values. Its goal is to create a mathematical formula that will give us a score between 0 and 1 as shown in Figure 9 [36].

A decision support tool called a decision tree employs a tree like graphical representation which is often used for explaining the decisions with their consequences, containing utility, resource costs, and outcomes from random events. It’s a tool by which an algorithm can be displayed.

In a decision tree, each class label is represented by leaf node and test results are represented as branches, while every internal node resembles a “test” property (for instance, whether a coin will fall on its head or tail) (decision determined by calculating all characteristics). The routes between root and leaf indicate the rules of classification [36].

A probabilistic classification model is created from previously evaluated attributes using a Nave Bayesian classifier, which is based on Bayes’ theorem. The filtration process of mail for antispam in which the filter is trained to extract the spam from legitimate mail automatically, represents its most common use. in the binary fashion [37, 38]. By doing this, it is possible to discern between spam and valid emails by matching each characteristic to terms found in both; the latter are then prohibited by a trained filter. As shown in Table 1, it consists of seven variables that mentioned earlier. Type of project was sample by number as (1= electrical, 2=water, 3= road, 4=residential) projects while cost symbol by (1 or zero) which mean the former that cost added where the later there is no cost added. The class on the other hand, represents the performance as 0 indicate low performance and 1 indicate a good performance.

Table 1. Subcontractor performance factors

Table 2. Sample data

Table 3. Correlation matrix

10.jpg

Figure 10. Relationship between three variables

11.jpg

Figure 11. Relationship between the last three variables

The data was gathered from 234 different projects and with the main featured, in the first step a relation between each variable and the performance was drawn.

The next step was to find the correlation between the variables. Table 2 shows a sample of the data that used in the paper. Each factor correlation with other were calculated as Table 3.

Figure 10 shows the relationship between the first three variables with all others and Figure 11 show the last three variables with all others.

After that, a correlation matrix was plot to see the connection between each variable and the class as in Table 2. Then, three algorithms were applied to see the accuracy of each one and to see how to get better performance. The results of the three algorithms are shown in Table 4.

Table 4. Algorithms accuracy

Entropy is simply our dataset’s uncertainty or a measure of disorder. Let’s say a group of pals chooses the Sunday movie they will see together. Everyone is now required to state which of the two movie options—”Lucy” or “Titanic”—they chose. Following everyone’s response, “Lucy” receives four votes, while “Titanic” receives five. Now, what movie are we watching? Given that the two films had about the same number of votes, isn’t it difficult to select only one? The fact that both films received the same number of votes and that we are unable to determine which to watch is precisely what we refer to as disorder. The decision tree classifies two class either yes or no, in our case it classifies the performance either good (1) or bad (0).

The sigmoid function is referred to as an activation function for logistic regression and is defined as:

$F(x)=\frac{1}{1+e^x}$      (1)

where, e=base of natural logarithms; value=numerical value one wishes to transform.

$Y=\frac{e^{(b 0+b 1 x)}}{1+e^{b 0+b 1 x}}$        (2)

where, x=input value; y=predicted output; b0=bias or intercept term; b1=coefficient for input (x).

In our case, there are many input variables such as time, cost, complete, change order, and deviation. Accordingly, Eq. (2) becomes:

$Y=\frac{e^{(b 0+b 1232+b 263.1+b 318+b 415+b 582)}}{1+e^{b 0+b 1232+b 263.1+b 318+b 415+b 582}}$

$Y=0.001$

$P\left(Y=y_i \mid \mathrm{X}=x_k\right)=\frac{P\left(\mathrm{X}=x_k \mid Y=y_i\right) P\left(Y=y_i\right)}{\sum_i P\left(\mathrm{X}=x_k \mid Y=y_i\right) P\left(Y=y_i\right)}$

$P\left(Y \cos \alpha+\cos \beta=2 \cos \frac{1}{2}(\alpha+\beta) \cos \frac{1}{2}(\alpha-\beta)=y_i x\right)=\frac{1}{1+e^x}$

When the total in the denominator is over all permissible values of the random variable Y, and ym indicates the mth possible value for Y and xk the kth possible vector value for X. Using the training data to estimate P(X|Y) and P(Y) is one method of learning P(Y|X). Then, we can use these estimates along with the previously mentioned Bayes method to find P(Y|X = xk) for each new occurrence of xk.

Table 4 illustrates that the average values of classification accuracy of Naïve Bayes, decision tree, and logistic regression as Naïve Bayes has achieved 0.1% higher average classification accuracy results in comparison to decision tree and about 2% than logistic regression.

For classification issues, logistic regression is employed. The Logistic regression problem’s outcome can only fall between 0 and 1. When probabilities between two classes are needed, such as true or false, 0 or 1, logistic regression can be applied. The foundation of logistic regression is the idea of Maximum Likelihood estimation, which makes use of the sigmoid function and yields a curve known as the S-curve or sigmoid curve. The primary requirement is that the data be linear [44]. The following are some advantages of logistic regression. First, a substantial association between the dependent and independent variables may be demonstrated via logistic regression. Secondly, the influence of variables assessed at various scales, such as the impact of price adjustments and the quantity of promotional activities, may also be compared using logistic regression analysis. This advantage aids in the process of removing and assessing the ideal collection of variables to be employed in the construction of prediction models by market researchers or data analysts. Third, in addition to being a classification model, the logistic regression model also yields probability-related data. to have a superior outcome Prior to employing logistic regression, each independent variable needs to have a valid value. Second, multinomial and categorical results may be accurately predicted using logistic regression. Third, because our input data is not between (0.1), there is no multicollinearity between the variables in the dataset [45-47], which lowers the accuracy.

Decision trees are the supervised classification method most frequently used. Learning and categorization are components of decision tree induction. Decision trees may be used in any area because these procedures are quick and easy to follow [48]. The following are some benefits of employing decision trees:

• Extremely low complexity: This method is simple to apply and doesn’t require a high level of statistical knowledge.

• Beneficial for data exploration: The decision tree approach is one of the quickest for producing or identifying new features, therefore it may also be used for data exploration.

• Requires less steps for data cleansing - In contrast, it remains unaffected by missing values and data.

• No limitations on data types: It has the ability to handle both category and numeric variables.

• Non-parametric approach: The decision tree method does not assume anything regarding the geographical distribution because it is non-parametric.

The drawbacks of decision tree algorithm:

• One of the main practical issues influencing the decision tree model is overfitting. However, overfitting issues can be reduced by trimming and imposing model parameter limits.

• Unsuitable for continuous variables - When classifying variables into different groups, the decision tree loses some important information [49].

Naive Bayes give the highest accuracy as this this algorithm has several advantages as follow:

• Handles both continuous and categorical characteristics;

• Is easy to understand and computationally efficient;

• Functions well with high-dimensional data;

• Works well with limited training datasets [50].

The comparison between the algorithms is as shown in Table 5:

Table 5. Comparison between Naive Bayes, decision tree and logistic regression [51, 52]

3.3 A system for the management of subcontractor

One of the main problems in the management of subcontractor is the weakens in the coordination between the main contractor and the subcontractor, therefore a system was built in order that the subcontractor make a daily, weekly or monthly report to avoid the arising of the problems. The main methodology shown in Figure 12.

The basic steps of the proposes system as follow:

1. Determine the subcontra or information

2. Identify the attributes for each subcontractor

3. Determine the projects activities

4. Determine each problem with its related description in order to quantify the amount of the problem

5. Determine the effect of each problem that occur in the projects

6. Compare the highest issue identified for each factor to each

7. Importance value for the most important issue for that factor

8. Then select the significant problems and send it as report to identify the preferred alternative, which provides the greatest amount of importance for the least amount of cost

12.jpg

Figure 12. System flowchart

13.jpg

Figure 13. System interface

14.jpg

Figure 14. Project description

Each node describes a function of the system as explained as in Figure 13 that explain the interface of the program that required a general description of the user, Figure 14 show the project description in term of cost, time, stuff, equipment and other information. The next step is to move to the lowest level which is the problem description as shown in Figure 15 in term of each problem effect and significant.

After project description, each problem encounter in the project will be described and then plotted.

The code that is used to program the app. Is as follows:

staff={‘Problems’, ‘Field’};

            if ~isempty(find(strcmp(selectedNodes.Text,staff),1))

                enableForm(app);

            else

              app.NameEditField.Value = ‘‘;

            app.sigEditField.Value = 0;  

            app.EmploymentEditField.Value = ‘‘;

15.jpg

Figure 15. Problem description

After problem description, it will be displayed in chart to make it more understandable in term of plot as shown in Figure 15.

The final step is to send a report to the owner that state all problems that occur during the projects which explained in Figure 16.

The application of this system can be used in general construction projects in order to supervise the subcontractor and make a contact between the main parties of the projects to make a solution in time to prevent the failures of the project.

16.jpg

Figure 16. Problem chart