intelligent project approval cycle for local government
DESCRIPTION
In e-government, decision makers need support in their decision processes that may vary from simple nature to complex one. Authorities desire an intelligent workflow for their multilevel approval cycle. In this paper, we propose to use Case Base Reasoning (CBR) for the approval of small projects in public sector. CBR is an artificial intelligence technique which efficiently exploits the past experience to find solution of new problems. The CBR engine maintains a repository of past cases. On a new project approval request, the proposed inference system matches similar historical cases and suggests a solution for the new project. The proposed methodology has been evaluated on a case-base of sample projects.TRANSCRIPT
Intelligent Project Approval Cycle for Local Government
Case-Based Reasoning Approach
M Kashif FarooqMalik Jahan Khan Shafay Shamail Mian M Awais
ICEGOV09 Partners
Intelligent workflow for multilevel project approval cycle
Application of CBR for approval of small projects in public sector
The CBR engine maintains a repository of past cases
On a new project approval request, the proposed inference system matches similar historical cases
Suggests a solution for the new project
Abstract
CCB Citizen Community Board –
Small CSO Civil Service Organization at local level
CCB proposes small local level projects related to social development or public service delivery
Projects may be small schools, health units, drinking water units, Roads and streets parks, vocational training centers, advocacy movements for society (public awareness)
Application Domain
Social Welfare Department (SWD) of local government receives these project proposals from local CCB
CCB has to fund raising up to 20% of total project cost to show public interest
If SWD approve the project, then SWD grants 80% amount of total cost of project
Application Domain
To provide support and automate the technical evaluation of the project by using CBR (Case Based Reasoning)
Scope of the Paper
Evaluation parameters may be grouped into the two clusters:
Class A: Objective Parameters
Can be evaluated by formula, rule or principle
Class B: Subjective Parameters
Can be assessed by experience
Evaluation Parameters
Nature of Project
Budget
Profile of CCB or CSO
Experience of CCB or CSO
Cost of Service
Class A: Objective Parameters
Need of Project
Socially Viable
Socio-economics
Political Support
Sustainability
Quality of Service
Class B: Subjective Parameters
CBR Based Proposed Approach for Project Appoval
Collection of n parameters as ith case of the case-base in the form of a vector as given in equation:
C represents a case and each parameter Pij represents defined parameters from the project approval dataset.
A case base CB containing m cases may be represented as given in equation:
Case Preparation
j
iniiiji PPPPC )..,....,( 21
k
mk CCCCCB )....,,( 21
There are many case retrieval methods to match the current case and a case in the case-base.
Some well known methods are Manhattan distance, Euclidean distance, Mahalanobis distance, Geometric similarity measures, and Probabilistic similarity measures
Case Retrieval and Reuse
Our empirical study suggests that Manhattan distance is the most suitable similarity measure for the domain of project approval cycle as our selected parameters to represent the relevant cases are of numeric nature.
It is used to retrieve matching cases from the case-base. It calculates the weighted sum of absolute differences
between the current case and any other case in the case-base.
This weight is set by the user or analyst. It is given by as
Where dij means distance between ith and jth cases with respect to all parameters
W represents weight. x is the current case while c is the historical case from CB
Manhattan or City distance
k
jkikkij cxWd
In first phase we implemented it on the projects related to the development of small health care units.
Experts finalized few critical parameters for the approval of projects.
We studied 50 cases and created a case library
Implementation
Table 1. Sample Case Data
# Parameters Health Care Projects
P1 P2 P3 P4
1 Need of Project (NP) 2.5 3 1.5 2.75
i Existing facility 3 2 0 2
ii Available Alternative 2 3 2 4
iii Capacity of existing and alternatives facility 2 4 3 3
iv Quality of existing and alternative service 3 3 1 2
2 Socially Viable (SV) 2.25 2.75
2.5 2.75
i Cultural conflict 2 3 2 2
ii Religious conflict 2 2 3 3
iii Awareness and literacy 3 2 3 4
iv Negative believes 2 4 2 2
3 Socio-Economics (SE) 3.3 3 4 3.6
i Affordability 3 3 5 4
ii Average income per person 4 3 4 4
iii Available low cost alternative 3 3 3 3
4 Technically Viable (TV) 3.25
3 3.5 3
i Availability of trained staff 3 3 2 3
ii Sustainability of trained staff 4 3 4 3
iii Availability of utilities (energy, supplies, communication, etc.)
3 2 4 2
iv Technical support for equipment 3 4 4 4
5 Political Support (PS) 2 3 3.3 3
i Political ownership 3 3 4 2
ii Political stability 2 4 3 4
iii Political conflicts 1 2 3 3
6 Sustainability (S) 3 3.6 4 3
i Financial sustainability 3 4 3 3
ii Legal sustainability 2 3 4 3
iii Institutionalization 4 4 5 3
7 Quality of Service (QS) 3 3.3 3 3
i Customer or citizen satisfaction 2 4 3 3
ii By social audit 4 4 3 2
iii By media trial 3 2 3 4
Seven macro parameters have been used to define project evaluation
One parameter Predicted Probability to predict the matching solution
Ci = (NP, SV, SE, TV, PS, S, QS, PP)
PS: Political Support S: Sustainability QS: Quality of Service PP: Predicted Probability
of project acceptance
RESULTS
NP: Need of Project SV: Socially Viable SE: Socio-
Economics TV: Technically
Viable
Leave-One-Out (LOO) has been applied as the cross validation method on a dataset of 50 projects
All of the data items were labeled, so it was supervised learning process
We used solution of one nearest neighbor for reuse phase
We adapted simplest revision mechanism which is suggested to pick the second nearest neighbor if the first one does not fit in
RESULTS
Metric Result
Accuracy 90%
AAE 0.011
RMSE 0.0376
RESULTS
We used three evaluation metrics to validate our results
50 iterations ? computed accuracy, root mean squared error
(RMSE) and average absolute error (AAE)
Our proposed approach is expected to provide benefits such as
quick and efficient decision making process with impartial, high quality and informed decisions
Current work involved pre-processing of data and did not deal with ambiguous input parameters, it would be very useful to deal with ambiguity and vagueness of the real data in future work
CONCLUSION AND FUTURE WORK