APPLICATION OF NEURAL NETWORKS FOR PRIOR

APPRAISAL OF STRUCTURAL FUNDS PROJECT PROPOSALS

Tadeusz A. Grzeszczyk

Institute of Production Systems Organization, Warsaw University of Technology, ul. Narbutta 85, 02-524 Warsaw, Poland

Keywords: Appraisal, Structural Funds, Neural Network

Abstract: The subject of present paper is to discuss the layout of conception referred to the use of artificial

intelligence methods (neural networks) for prior appraisal of project proposals to be submitted by Polish

enterprises to European Union in order to get financial assistance for investments from the EU structural

funds and the state budget. The experiments are limited to prior appraisal of the projects submitted only, as

their practical execution may begin not earlier than on the 1

May 2004 (enlargement of European Union).

Author of the present paper discusses the method referred to appraisal of project proposals submitted by

enterprises. The method is related to review and acceptance of expenditures for investments co-financed by

European Regional Development Fund. The author formulates conception for implementation of appraisal

principles which could be considered as element of review and acceptance of expenditures according to

Commission Regulation 1685/2000.

1 INTRODUCTION

Advanced process of transformation which takes

place in Poland is intended to enable accession to

European Union in near future. It offers exceptional

unique chance for the enterprises to make use of

European funds which are to be granted in still wider

and wider extent. For the time being, so-called pre-

accession funds such as PHARE, ISPA, SAPARD

and some others may be mentioned in this

connection. On acquiring membership and full

access of Poland into European Union, much higher

European funds will be potentially available, among

the others, the structural funds.

However, effective use of potentially available

funds will require well-prepared, qualified staff for

Polish enterprises. It is necessary as well to supply

the enterprises with relatively simple (for easy

application) methods of appraisal, referred to

investment projects submitted by them. In particular,

the small and middle-size enterprises will have

appraisal problems because of rather limited funds

for expert advice. In this connection, author of the

present paper has worked out and submits herein, the

conception referred to application of neural

networks for prior appraisal of project proposals

while contending for European funds.

The main task of the present research project is

to explore the basic methods of appraisal related to

feasibility and effectiveness of development

programmes which are to be financed by EU

structural funds. Up to now artificial intelligence

methods have not been applied for such purpose.

Author of the present paper is going to compare the

hitherto prevailing methods to the methods based on

neural networks. The experiments will be held to

prove possibility as well as need for application of

neural networks within the process of project

appraisal. The experiments are limited to ex-ante

evaluation (prior appraisal) of the projects

submitted, as their practical execution may begin not

earlier than on the 1

May 2004. As the subject for

experiments, descriptions of potential projects

(selected by survey) and intended for financing by

ERDF means (European Regional Development

Fund) will be considered.

Up to now, there have not been notable

experiences with management of development

projects financed by EU funds in Poland. Some

attempts to apply artificial intelligence i.e. neural

networks for the above purpose, are considered as

something new. However, neural networks should

increase set of useful tools used for appraisal of

projects to be financed with EU funds and seem to

be a good supplementation of the methods used so

far.

501

A. Grzeszczyk T. (2004).

APPLICATION OF NEURAL NETWORKS FOR PRIOR APPRAISAL OF STRUCTURAL FUNDS PROJECT PROPOSALS.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 501-504

DOI: 10.5220/0002605505010504

 SciTePress

2 APPRAISAL OF STRUCTURAL

FUNDS PROJECT

In the procedure of appraisal referred to applications

for financial support of investment projects from

European funds, evaluation of documents (subjected

before to unification) and operation programmes

should be carried out in three implementation stages

as follows:

• operation programme is prepared and

supplemented (if required) together with

preliminary evaluation, prior to the beginning of

its implementation; so-called: ex-ante

evaluation,

• evaluation is carried out in the middle of

implementation period: so-called mid-term

evaluation,

• evaluation is carried out after the end of

implementation period: so-called ex-post

evaluation.

Prior appraisal of the project proposals can be

based on the analysis of strengths - weaknesses -

opportunities - threats (so-called SWOT).

Preliminary analysis of appraisal referred to

effectiveness of the project, should be done with

taking into consideration such elements as socio-

economical situation in general, situation on the

market, competitiveness and innovativeness. The

variables applied for appraisal of the project should

meet the following criteria:

• pertinence - the variables should be harmonized

with character of the project and with

anticipated effects of its implementation;

• measurability - the variables should be

expressed by means of numerical values;

• credibility - definition of any variable should

enable possibility of its verification;

• accessibility - the data should be easy to obtain.

The above mentioned properties of variables are

taken into account in the process of preparation and

implementation of appraisal models based on the

methods of artificial intelligence.

3 NEURAL MODELS

Principal properties of neural models are important

to determine their effectiveness. Application of

dependences appearing in neural networks, does not

involve formulation of assumptions which are very

difficult for checking. They are characteristic

because of their ability for approximation of optional

nonlinear dependences. They make possible

generalization of learning, from training (historical)

data towards the new data. Formation of neural

model is based on analysis of available historical

data. In result, the main dependences in phenomenon

being investigated, can be esteemed (with the use of

model). These models are useful in case, the

research worker does not know the rules

characteristic for arising of dependences analyzed.

They are particularly useful for description of

variable, complex socio-economical phenomena.

Application of neutral networks requires

adequate preparation related to significant number of

historical data, according to the character of

variables and type of network to be used. However,

this is connected with work and expenditure

necessary to estimate neural model. In this case, it is

most often assumed that number of samples for

learning should be minimum 10 to 20 times higher

than number of weights in the network. Direct and

substitutable dependence between number of

samples from the learning set and precision of

results was not stated empirically. It should be

noted, however, that in extreme cases, at small

number of data (i. e. comparable to the number of

weights), the network in unable to generate proper

results.

Effectiveness of neural networks in appraisal

procedure is expressed as description and analysis of

optional dependences and their generalization. Due

to the above, the neural network is able to give

correct answer to a formulated problem, on the basis

of already transformed input data. In this way

information about regular events from the past is

available. Functioning of the model is based on the

assumption that the information collected is

typically representative for any other data which

might appear. Highly desired property of neural

network is its ability for formulation of proper

answer, even for those input data which are not

included within the set of already collected

information. Applicability of one-way neural

networks for appraisal of investment projects, results

from the fact that socio-economical phenomena have

nonlinear character. These networks are able to

perform approximation of optional nonlinear

dependences and their generalization. They are

adaptable and can help for description of

dependences changing in time. Therefore, during the

input of new information, further network learning

process takes place. Sometimes, the learning process

is limited to small amendments only or to taking into

account transformations of the real system. One-way

neural networks can be particularly useful for

solving the problems of different types of markets

(e.g. capital market, labour market, markets of goods

and services). One-way neuron networks can serve

as well for analysis of problems related to

functioning of enterprise, also while applying for

financial assistance from European funds intended

ICEIS 2004 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

502

for investment projects and for estimation of

anticipated profits.

4 METHODOLOGY AND

EXPECTED RESULTS

The main task of the scientific-research project is to

analyse the methods of appraisal referred to

feasibility and effectiveness of development

programmes to be financed by EU structural funds,

as well as checking possibility and real purpose for

application of new methods based on artificial

intelligence. Satisfactory training of beneficiaries

and public administration for absorption of EU funds

are considered as essential for economy of Poland.

Significant socio-economical losses can be involved

if they are unable to fully profit of EU structural

funds potentially available beginning with May

2004.

Research works undertaken by the author are

going to be limited to so-called ex-ante (prior to)

methods of project appraisal to be carried out before

the beginning of development projects. Directives

and recommendations of European Commission pay

particular attention to a role of a.m. ex-ante

evaluation in order to improve quality of the projects

submitted. The potential beneficiaries are required to

prepare (among the others): feasibility study, ex-ante

project appraisal and to indicate some alternative

solutions. The preliminary ex-ante evaluation

(appraisal) will give some arguments for discussion

(between the persons or institutions submitting the

projects and the experts) to get more precisely

defined development programmes. The other

advantage of ex-ante project appraisal is that it

makes the authors of development programmes to

think them over again then introduce some necessary

improvements as to their effectiveness and

conformity with directives of European

Commission. The quality and quantity factors

formulated during ex-ante project appraisal, can be

also useful for evaluation and monitoring at further

stages of project execution and monitoring.

The research will consist in experiments with

methods (according to directives of European

Commission) which have been applied so far in

Europe, as well as with methods of project appraisal

based on artificial intelligence - neural networks.

Among the others, the software „Statistica Neural

Networks” by Statsoft will be used. This is relatively

simple programme which serves for simulation of

neural networks and for construction of system

including different types of networks. Easy access to

graphics and statistical tools as well as interface

being very convenient for the user, make possible

quick and efficient interactive analysis of the data.

As a starting point, identification of project

appraisal methods applied nowadays (in the

countries being members of European Union) for

appraisal of projects to be financed by EU funds, is

considered in accordance with directives of

European Commission. Then, the neural networks

will be subjected to research as a tool convenient for

analysis of socio-economical data (as described in

corresponding literature). Then, possibility of

potential application of neural networks for

management of development projects to be financed

by EU structural funds will be analysed. Next, the

comparison of project appraisal methods applied so

far with the project appraisal methods based on

neural models is to be done. Faults and features of

all a.m. solutions will be examined. Within the next

stage of research, the author will focus on making

experiments with different types of neural networks

which could be used for project appraisal. The

experiments will be limited to ex-ante appraisal of

the projects submitted, as they may be put into

practical execution after 1

May 2004, not earlier.

The experiments will be performed on the projects

potentially predicted for financing by European

Regional Development Fund (ERDF), selected by

means of survey. These projects are prepared

(among the others) on the basis of charts „Internet

System of Evidence for Charts for Projects Intended

for ERDF” (http://isekp.mg.gov.pl). Data base for

these projects can be found in Ministry of Economy,

Labour and Social Policy of Poland. It is registered

as confidential. However, the author hopes to

receive all the necessary data for scientific

elaboration from the persons directly connected with

preparations of the projects. Such are preliminary

conceptions. Up to the beginning of June 2003,

about 4300 potential projects have been submitted.

Methodological tasks of research do not require

representative samples (as for survey); only

typology and differentiation of subjects and areas

can be important. The research and analysis of about

200 projects is predicted.

5 CONCLUSIONS

This is advisable to make research for alternative

methods of appraisal in relation to currently applied

ones. The methods having connection with artificial

intelligence should be investigated and applied first.

Author of the present paper recommends application

of neural networks. Neural networks are, for the

time being, decisively most often used tools (within

the range of artificial intelligence) for analysis of

APPLICATION OF NEURAL NETWORKS FOR PRIOR APPRAISAL OF STRUCTURAL FUNDS PROJECT

PROPOSALS

503

socio-economical data. The operation of neural

networks is based on their self-learning ability or

their supervised education; in result models of

events to be discussed are originated with the use of

algorithms of stochastic type.

Neural networks are recommendable for their

following desired qualities:

• They are non-linear and non-parametric, no

form of function allows for shaping the model

of input-output relations.

• Any assumptions referred to forms and

parameters of random variables distribution are

not required.

• They are resistant to disturbances which appear

in real systems.

• Application of neural networks makes possible

acquiring of sufficient additional knowledge

necessary for appraisal as well as in selection of

information essential for appraisal purposes and

elimination of unimportant factors.

• In case of the classical statistic methods being

applied, it is necessary to determine dependence

between the active (explaining) and passive

(being explained) variables; reasons for such

configuration should be given further on. All the

above steps are not required while applying

neural networks.

• The characteristics formulated above should be

considered also as some limitation for applying

neural networks, as the process of achieving

results cannot be supported by any system of

variables.

The aforementioned qualities of the systems

based on artificial intelligence methods, may be used

in appraisal systems competitive in relation to these

which have been applied up to now. The new

appraisal system should, if possible, make use of

neural networks.

New developments in data processing techniques

involve intensive research for new methods referred

to support of appraisal process based on artificial

intelligence. However, the following limitations in

application of instruments discussed herein, should

be taken into account:

• Significant computation expenditure required.

• In view of the above, application of traditional

regressive equations should be reasonable for

appraisal, if small numbers of historical data are

concerned.

• Relatively simple application of neural

networks sometimes results in wrong suggestion

that the user does not need to analyse quality of

the model prepared - what is usually done if

traditional statistical methods are applied.

• The only method of verification referred to

results of analysis can be their comparison to

the real existing data.

Thorough analysis of the above characteristics

referred to the methods classified as artificial

intelligence (rights and wrongs) leads to the

conclusion that their application may result in

reduction of appraisal errors - in relation to the other

traditional statistical methods.

The conception outlined herein (including tools)

can be applied in enterprises which are going to

apply for financial assistance from EU pre-accessive

funds now and structural funds after becoming a

member country of European Union The research

carried out by author of the present paper, referred to

application of neural networks for preliminary

appraisal of structural funds project proposals, can

find also wider application involved by some other

problems e.g. implementation of the other EU funds

included in Polish National Development Plan -

within the range of Community Support

Framework.

REFERENCES

European Commission, 2000. Directorate General for

Regional Policy and Cohesion. Working paper 1.

Vademecum for Structural Funds Plans and

Programming Documents. Brussels.

European Commission, 2000. Directorate General for

Regional Policy and Cohesion. Working paper 2. The

Ex-ante Evaluation of the Structural Funds

Interventions. Brussels.

European Commission 2000. Directorate General for

Regional Policy and Cohesion. Working paper 3.

Indicators for Monitoring and Evaluation: An

Indicative Methodology. Brussels.

European Commission, 1999. MEANS Collection.

Evaluating Socio-economic Programmes. Volume 1 –

6. Luxemburg.

Grzeszczyk T. A., 2000. Rough Sales Forecasting System.

Informations Systems – Research, Teaching and

Practice. Proceedings of the 5th UKAIS Conference

University of Wales Institute. Cardiff 26 – 28 April

2000.

Grzeszczyk T. A., 2002. Rough Sets Theory for Sales

Forecasting System. In: New Challenges And Old

Problems In Enterprise Management. Edited by T.

Krupa. WNT. Warsaw.

http://isekp.mg.gov.pl

http://www.mg.gov.pl

http://www.statsoft.com/stat_nn.html

ICEIS 2004 - ARTIFICIAL INTELLIGENCE AND DECISION SUPPORT SYSTEMS

504