Intelligent Decision Support System for Construction Project Monitoring

Transcription

1 Intelligent Decision Support System for Construction Project Monitoring Muhammad Naveed Riaz Faculty of Computing Riphah International University Islamabad, Pakistan. Abstract Business Monitoring is a complex task and it has been noted that most of the reporting and analysis time is being spent on collecting data from the various systems. Over the past decade, a lot of research has been reported on Decision Support Systems (DSS) used in many fields. To improve the decisionmaking ability of an enterprise in construction management, information technology is being applied in each step of construction management. The problem is to organize and analyze the data in construction management to obtain quick analysis and decision support results. Various Data mining techniques have been used for clustering of data by using case examples. In this research we have applied Learning Vector Quantization (LVQ) to classify projects in one of the given categories and conducted a comparative analysis by using standard algorithm. A number of case examples have been used to verify the results and to obtain a comparison between various methodologies. Keywords-component; Decision Support Systems; Learning Vector Quantization; Business Intelligence; I. INTRODUCTION In business, Decision Support Systems (DSS) are usually employed for the analysis of data to find solutions or strategies that are useful in effective decision making. Knowledge base is intelligent component used in IDSS to suggest some useful activity in the human decision making process, that can make DSS supporting to decision makers [1]. One of the most important features which derive construction from other processes is the difficulty of the processes with a number of stages includes in construction which needs to be properly managed and addressed [2]. In most of the construction projects it is hard to supervise the projects without an effective project monitoring system because the projects may be spread over large area, needed strict time lines and multiple decision makers present in different locations. There may be thousands of projects running simultaneously and each project has its own timeline which has to be monitored properly and timely. The research and development on intelligent decision support systems for construction management can be done by using the Data Mining Techniques (DMT) and neural networks which offers possible solutions to these problems. The objective of this research is to design such an intelligent decision support system for construction project monitoring that is helpful to provide a strong base for organizational decision makers to Syed Afaq Husain College of Computer Science & IT King Faisal University Saudi Arabia. drafaqh@gmail.com take the right decision in a timely manner for the enhancement of business. II. CURRENT WORK The ongoing research on IDSS suggests that DSS are rarely applied at organization level and very occasionally applied in a construction branch [3]. Data Mining (DM) is one of the most renowned tool which is used in BI for the extraction of hidden information from the databases [4]. The data mining phase covers selection and application of DM techniques, which are initialization and calibration of useful parameters to find optimal values [5]. The intelligent decision support system for bridge monitoring is based on case-based reasoning which needs visual inspection of data and non destructive testing [6]. The DSS are rarely used in Neural Networks and has been successfully applied to a wide range of real-world construction applications, such as (Mohan, 1990): cost management, quality control, signal processing, credit rating, sales forecasting, modeling, quality control, portfolio management, targeted marketing and education, finance, etc [3]. The LVQ (Linear Vector Quantization) network with fuzzy feedback function algorithm can not only unfold the features of water resource requirement after the historic data analysis, but also generate new input data with some reasonable problem resolves, which makes the algorithms with feedback and evolvement scheme [7]. Various Clustering algorithms have been used now a days but no comparative study for their utilization in construction management is available. LVQs and support vector machines have been utilized for effective clustering in various fields but their performance for data mining in construction management needs to be evaluated against other standard techniques. DMT can be applied on the historical data stored in database to identify similar problems and classify the problem into relevant category based on its similarity. A web based application on intelligent decision support system for construction project monitoring (IDSSCPM) was designed and developed to cater to the problematic data sets of construction projects into the relevant category which are known as problem scenarios e.g stuck projects, latest or ongoing projects, delayed or behind schedule and completed projects etc. The Figure1 is composed of stuck projects running in the field area. The stuck projects are those projects which are /12/$ IEEE

2 stuck in the field due to some reason which includes Work not Started yet, No Activity at Site, Land Issue and Contract Terminated. The construction management requires the dynamic analysis of projects by visual representation and timely decision making is needed to complete these projects within the required time duration. Figure 1 Stuck Project. III. PROPOSED METHODOLOGY In order to address the defined problem scenarios mentioned above, an Intelligent Decision Support System for Construction Project Monitoring (IDSSCPM) has been proposed for constantly monitoring the projects of construction. If the user wants an analysis of the problem scenario, the IDSSCPM system shall perform data mining techniques on the historical data stored in the database to identify similar problems and categorize the problem into the relevant group. The proposed system shall identify the problems if and when they occur and raise flags as discussed previously. The proposed system architecture is shown in Figure 2 below. the most complicated and difficult part lies in knowledge gain, inference and natural language processing system [8]. The latest research on Artificial Neural Networks (ANN) indicates that ANN is used to enhance the capabilities of the intelligent decision support system [4]. Data mining techniques provides Clustering as a method for grouping of similar data. The LVQ algorithm is used to form the clusters. The LVQ is a type of competitive learning neural network such as the Self- Organizing Map (SOM) algorithm for unsupervised learning with the addition of connections between the neurons. LVQ network belongs to the competition neural network which includes the input layer, competition layer and linear output layer [9]. In the first phase the LVQ neural network does not need to normalize the input vector and in the second phase it only need to calculate the distance between the input vector and the competition layer directly [10]. The learning algorithm can analyze and clusters the construction data (Case Examples) to find the trend analysis which are further helpful for decision making. The objective of the study is to provide a system based on the neural network (LVQ algorithm) for clustering of data which provide the detail analysis of construction projects. The LVQ algorithm is selected and implemented using MATLAB which is used to extract the Clusters. The modified LVQ for update of weight vector used is: w j (new)=w j (old) + (α (Iteration) * (x - w j (old))) The parameters and values taken for the LVQ algorithm are as fallow: 1. Weight Initialization Method = SOM 2. Number of Training Vectors = Learning Rate α = Total Number of Learning Iterations = 100 The algorithm performs clustering on the construction projects to form clusters. The results of three case studies are applied e.g. stuck projects, behind schedule or delayed projects, progress satisfactory or completed projects. The analysis of the projects is shown graphically according to progress percentage of projects and the number of projects in each cluster. The analysis is also shown as districts wise and construction activity wise for trend analysis of different projects. Microsoft Clustering algorithm is also used to form the Clusters for the comparative analysis with LVQ Clusters. IV. RESULTS & ANALYSIS: Figure 2 Architectural View of Proposed Methodology. DSS with neural network provides a new dimension for the development of DSS with the traditional AI, in which A. Clustering of Stuck Projects: Table 1 shows the clusters output extracted from the data sets of stuck projects by using LVQ algorithm. The 8th attribute shows the extracted clusters formed from the data sets. From Table1 the numeric value 1 represents the progress percentage between 0 and 5, the numeric value 2 represents the progress percentage between 6 and 20, the numeric value 3 represents the progress percentage between 21 and 40 and similarly for value 4 and 5 as shown in Table 1.

3 Figure 4 shows the stuck projects analysis with respect to the districts but with separate clusters e.g. the district Abbot shown as numeric value 1. Cluster1 contains the number of stuck projects 3; Cluster2 contains the number of stuck projects 2 and so on in Figure 4. Table 1 Clusters Output of Stuck Projects Figure 3 explains the characteristics of the clusters formed by LVQ algorithm. It shows the progress percentage of stuck projects ranges for each cluster produced e.g Cluster1 ranges from 0 to 5 % of progress where as Cluster5 ranges from 64 to 84 % progress. Maximum number of stuck projects exists in Cluster1 which are 30 and similarly 4 stuck projects fall into Cluster 4 which is the minimum number. Figure 4 Clusters of Stuck Projects w.r.t. Districts Analysis of stuck projects w.r.t. Construction Activity: Table 3 shows the clusters formed by construction activity, from the stuck projects e.g in construction activity Contract Terminated Cluster1 contains the number of stuck projects 2, In Cluster4 the number of stuck projects is 0 which is the least number means no project is stuck. In construction activity No activity at Site, Cluster5 contains the number of stuck projects 11 which are highest in number. Table 3 Analysis of Stuck Projects w.r.t Construction Activity Figure 3 Clusters of Stuck Projects w.r.t Progress Percentage Table 2 shows the District wise clustering formed for the stuck projects e.g in district Abbot Cluster1 contains the number of stuck projects 3, Cluster2 contains the number of stuck projects 2 and Cluster5 contains the number of stuck projects 11, which are highest in number. The Cluster2 and Cluster3 contain the minimum number of stuck projects 2. Figure 5 shows the stuck projects analysis with respect to the construction activity with separate clusters e.g. the construction activity Contract Terminated shown as numeric value 1 in Figure 5. The Cluster1 contains the number of stuck projects 3; Cluster2 contains the number of stuck projects 2 and so on. Table 2 Analysis of Stuck Projects w.r.t. Districts. Figure 5 Clusters of stuck projects w.r.t. Construction Activity

4 B. Clustering of Delayed Projects: Table 4 shows the clusters output extracted from the data sets of delayed projects by using LVQ algorithm. The 8th attribute shows the extracted clusters formed from the data sets. From Table4 the numeric value 1 represents the progress percentage between 0 and 14, the numeric value 2 represents the progress percentage between 15 and 30, the numeric value 3 represents the progress percentage between 31 and 55 and similarly for value 4 and 5 as shown in Table 4. number exists in Cluster3, while Cluster1 contains minimum number of delayed projects 5. Table 5 Analysis of delayed projects w.r.t. Districts Figure 7 also shows the delayed projects analysis with respect to the districts with separate clusters e.g. the district Abbot is shown as number 1 in Figure 7. The Cluster1 contains the number of delayed projects 5, Cluster3 contains the number of delayed projects is 10, which is the maximum number of projects in this cluster. Table 4 Clusters Output Figure 6 explains the characteristics of the clusters formed from the LVQ algorithm. Figure shows the progress percentage ranges for each cluster produced e.g. Cluster1 ranges from 0 to 8 % progress, where as Cluster5 ranges from 80 to 100 % progress. Maximum number of delayed projects exists in Cluster1 which are 19 and similarly 7 delay projects fall into Cluster5 which is the minimum number. Figure 7 Clusters of Delayed Projects w.r.t. Districts Analysis of delayed projects w.r.t. Construction Activity: Table 6 shows the construction activity wise clustering formed for the delayed projects e.g. In construction activity No activity at Site, Cluster1 contains 19 delayed projects, which are maximum number of projects in Cluster1. In Cluster5 the number of delay projects is 7 which is the minimum number of projects to be delayed. Figure 6 Clusters of Delayed Projects w.r.t Progress Percentage Table 5 shows the District wise clustering formed for the delay projects e.g in district Abbot Cluster1 contains 5 delayed projects, Cluster2 contains 8 delayed projects and Cluster3 contains 10 delayed projects. Thus the highest Table 6 Analysis of delayed projects w.r.t. Construction Activity Figure 8 shows the projects analysis with respect to the construction activity e.g. in construction activity No activity at Site, Cluster1 contains the number of projects 19, Cluster4 contains the number of projects 12 and Cluster2 contains 8 projects.

5 Table 8 shows the district wise clustering formed for the completed projects, indicating the number of projects that are completed in each district e.g. in district Abbot, Cluster1 contains the number of projects 4 and Cluster5 contains the number of projects 24 which are maximum in number, Cluster4 contains minimum number of completed project 0. Figure 8 Clusters of Delayed Projects w.r.t. Construction Activity C. Clustering of Completed Projects: Table 7 shows the clusters output extracted from the data sets of completed projects by using LVQ algorithm. The 8th attribute shows the extracted clusters formed from the data sets. The numeric value 1 represents the Cluster1 and numeric value 2 represents Cluster2 and so on. The clusters are shown with respect to project completion. Table 8 Analysis of Completed Projects w.r.t. Districts Figure 10 shows the completed projects analysis with respect to the districts but with separate clusters e.g. in district Abbot, the Cluster1 contains the number of completed projects 4, Cluster2 contains 1 completed project. Table 7 Clusters Output of Completed projects Figure 9 shows that how the completed projects data sets are categorized into the clusters. These clusters are formed on the basis of total number of modules for each project which is completed. From Figure 9, the most of the completed projects belong to Cluster5 which are 56 in number, where as less number of completed projects belongs to Cluster4 which are 0 in number. Figure 10 Clusters of Completed Projects w.r.t Districts. Table 9 shows the construction activity wise clustering formed for the completed projects e.g. in construction activity Completed, Cluster1 contains the number of projects 4, whereas Cluster2, Cluster3 and Cluster4 contain no project. In Construction activity External development and final finishes, Cluster5 contains the maximum number of completed projects 50. Table 9 Analysis of Completed Projects w.r.t. Construction Activity Figure 9 Clusters of Completed Projects Figure 11 also shows the completed projects analysis with respect to the construction activity but with separate clusters e.g. the construction activity Completed shown as numeric value 1 in Figure 11, Cluster1 contains the number of completed projects 1, Cluster2 contains the number of completed projects 0.

6 and fine tuning of the learning algorithm. It is concluded from the results that the Clusters extracted from the LVQ are more flexible and dynamic then the results extracted from Microsoft Clustering. So, IDSSCPM aids the decision maker in effective decision making by providing them useful information, by dynamic analysis of the projects. The results achieved so far have been encouraging for decision making. However, there is a need to extend the work to include prediction and forecast of projects as well. REFERENCES Figure 11 Clusters of Completed Projects w.r.t Construction Activity V. COMPARATIVE ANALYSIS: The Microsoft clustering algorithm is selected for clustering of data sets for comparative analysis. The same data sets of stuck projects have been processed with Microsoft Clustering (BI Tool). Figure 12 Construction Activity wise Cluster Diagram. The clusters extracted from Microsoft Clustering are analyzed by the construction activity at which stage the projects are stuck, which provides the decision maker an effective analysis of projects as well. The Figure 12 shows the projects at stage of Contract Terminated mainly exists in Cluster5 and Cluster8. VI. CONCLUSION & FUTURE WORK The results obtained from LVQ algorithm have been compared with those obtained from the Microsoft Clustering. The Clusters extracted from Microsoft clustering are 10 where as the Clusters extracted from the LVQ algorithm is 5 which shows more flexibility and provides useful information. The LVQ Clusters obtained from the data sets is shown according to progress percentage of projects and the number of projects in each cluster. The LVQ results are flexible as (the clusters are shown by physical progress of projects, Districts wise Clusters, Construction Activity wise Clusters), and efficient because there is greater flexibility in the number of Clusters [1] Yang Bao, LuJing Zhang, Decision Support System Based on Data Warehouse, World Academy of Science, Eng. and Technology, [2] Sigitas Mitkus, Eva Trinkuniene, Decision Support in analysis of Construction Contracts, The 25 th International Symposium on Automation and Robotics in Construction. ISARC June, Vilnius, Lithuania. pp DOI: /isarc [3] A. Kaklauskas, E.K. Zavadskas and V. Trinkunas, A multiple criteria decision support on-line system for construction, Engineering Applications of Artificial Intelligence, Volume 20 Issue 2, March Pergamon Press, Inc. Tarrytown, NY, USA, pp DOI: /j.engappai [4] Maqbool Uddin Shaikh, Saif Ur Rehman Malik, Mohammad Ahsan Qureshi and Sarah Yaqoob, Intelligent Decision Making Based on Data Mining using Differential Evolution Algorithms and Framework for ETL Workflow Management In Proceedings of the IEEE 2010 Second International Conference on Computer Engineering and Applications - Volume 01, March 19-21, Washington, DC, USA, pp DOI: /ICCEA [5] Luan Ou and Hong Peng, Knowledge and Process Based Decision Support in Business Intelligence System. In Proceedings of the First IEEE International Multi-Symposiums on Computer and Computational Sciences, June Washington, DC, USA, pp DOI: /IMSCCS [6] Yin zihong, and Li yuanfu, Intelligent Decision Support System for Bridge Monitoring Proceedings of the 2010 IEEE International Conference on Machine Vision and Human-machine Interface, April 2010, Washington, DC, USA, pp DOI: /MVHI [7] Jian Wang, Yuanyuan Zhang, Research on Prediction of Water Resource Based on LVQ network, In Proceedings of 2011 International Conference on Electrical and Control Engineering (ICECE), Sept Yichang, China. pp DOI: /ICECENG [8] Kai Li, Zhonghua Xu, Baoqin Wang. Research of Intelligent Decision Support System based on Neural Networks, In Proceedings of the 2008 Second IEEE International Conference on Genetic and Evolutionary Computing, Sept Washington, DC, USA, pp DOI: /WGEC [9] Tao Xu Research on Sensor Fault Diagnosis Method based LVQ Neural Network and Clustering Analysis, In Proceedings of the 7 th IEEE World Congress on Intelligent Control and Automation June 25-27, Chongqing, China, pp DOI: /WCICA [10] Yao Xiao, Le Lei, Research on Comparison of Credit Risk Evaluation Models Based on SOM and LVQ Neural Network, In Proceedings of the 7 th World Congress on IEEE Intelligent Control and Automation, June 25-27, Chongqing, China, pp DOI: /WCICA