Soft Computing Models In Mode Choice
Keywords (Original Language):
Author Name | University of Author |
---|---|
Abstract (2. Language):
The aim of this study is to use soft computing methods
for modelling the mode-choice in urban passenger
transportation. For this purpose four softcomputing
models were developed and compared
with the conventional binary logit model by using
the data collected in the Transportation Master Plan
of Eskisehir (EUAP) in 2002. Travel survey data
were classified according to the income level of the
households: high, medium and low income. For each
income group two types of mode-choice models were
developed. 1. Conventional binary logit model. 2.
Soft computing models such as artificial neural networks
(YSA), pure fuzzy logic (BM) and neuro-fuzzy
logic (SB). Travel time and travel cost were used as
the two main parameters in all models. Mode-choice
was made between car and public transport. The
analysis covered the following issues; 1) Identifying
the best common performance measures to compare
the binary logit and soft computing mode-choice
models. 2) Calculation of performance measures. It
has been shown that soft computing models; especially
Sugeno type neuro-fuzzy models give better
estimates for predicting the mode-choice of the sample
data. It has been also shown that R2 could be a
sufficient measure to compare the performance of
the conventional and soft computing mode-choice
models. Finally, it has been shown that threshold
values that are not equal to 0.50 could improve the
performance of some mode-choice models.
Production and demand processes are fundamental
elements of economics. After industrial revolution,
human beings started to supply their needs instead
of self-agricultural production. In this dynamics,
minor and major choices shaped demand and production
phases. At the side of the goods or service
providers (car company or a municipality which
gives transportation service), prediction of choices
or factors affecting these choices are very important.
The ability to accurately predict the future is fundamental
to many decision processes in planning such
as scheduling, purchasing, strategy formulation,
policy making, and supply chain operations. Most
firms seek to maximize their market share. For that,
they need consumers that are on target, their behavior
structure.
All operational models for predicting individuals
(travellers) choices are based on behavioral principle called "utility maximization." According to utility
maximization principle, there is a mathematical
function U that is called utility function, whose numerical
value depends on attributes of the available
options and the individual (causal relationship).
The utility function has the property that its value for
one option exceeds its value for another, if and only
if the consumer prefers the first option to the second.
The consumer chooses the most preferred option,
which is the one with the highest utility-function
value (rational man).
The correct utility function may differ from that used
by the analyst due to the omission of a variable that
influences mode choice, measurement error, variations
in preferences among individuals, or all of
these. In each case the correct utility function U, can
be written as the sum of the utility function specified
by the analyst, V and an error term,ε .
Soft computing techniques such as neural networks
(ANNs), fuzzy logic (FL) and genetic algorithms
(GA) are very powerful modelling tools for transportation.
Zadeh (1965) introduced fuzzy set theory as
a general approach to express the different types of
uncertainty inherent in human systems. Artificial
neural networks (ANNs) are computing models for
information processing and pattern identification.
They grow out of research interest in modeling biological
neural systems, especially human brains.
Data for modelling was collected from Eskisehir
City in Turkey. Data sample was splited into three
parts according to income level such as high, medium
and low. For medium income group, two types
of modelling were structured. The first is a traditional
mode choice model consisting of binary logit
models. The second are soft computing models consisting
of artificial neural networks, pure fuzzy logic
and neuro-fuzzy models. Two main parameters were
used in all models are travel time and travel cost.
Choice set was formed of two choices, mass transit
and car. End of the analysis; neuro-fuzzy models
have the best prediction scores. In small sample size,
logit models do not give acceptable prediction
scores. Anyway, all logit models fail in all statistics
tests.
Bookmark/Search this post with
Abstract (Original Language):
İnsan davranışının matematik modellemesinin geldiği son aşama, insanı bitmek tükenmek bilmeyen
ve tatminini ençoklayacak şekilde davranan bireyler olarak tanımlamasıdır. Bu tanıma göre geliştirilen
ekonomik seçim modelleri, yararı en çok olan seçeneğin seçileceğini öngörür. Bu, ekonomide
yarar teorisi olarak anılır. Seçim modelleri deterministik ve rastgele yarar modelleri olarak ikiye
ayrılır. Deterministik yarar modelleri(DYM); basitçe, kullanıcı ve kullanıcıya sunulan seçenekelere
ait özelliklerin bileşimini gösteren bir fonksiyona göre en büyük yararı üreten seçeneğin kullanıcı
tarafından seçileceğini öngören modellerdir. Rastgele yarar modelleri (RYM) ise; deterministik yarar
teorisine rastgele bir hata teriminin eklendiği modellerdir. Bugün ulaştırma türü seçiminin modellemesinde,
en çok kullanılan modeller RYM’nin bir türü olan logit modellerdir. Gerek RYM gerekse
onun alt türleri olan modeller istatistik modellerdir ve varsayımlarına uygun olmayan koşullarda
bunların tahmin başarımları oldukça düşer. Esnek hesaplama modelleri olan yapay sinir ağları
(YSA), bulanık mantık (BM) ve genetik algoritmalar (GA) kullanılarak, geleneksel tür seçimi
modellerinin başarısız olduğu gözlemlerle, daha yüksek tahmin başarımları elde edilebilir. Bu çalışmada,
Eskişehir orta gelir grubu için geleneksel (logit model) ve esnek hesaplama yöntemleri
(YSA, BM ve Sinir-Bulanık) kullanılarak geliştirilmiş ulaştırma türü seçim modelleri incelenmiş ve
başarımları karşılaştırılmıştır. Özellikle Sinir-bulanık modeller, diğer bulanık modellere göre daha
iyi tahminler yapmıştır. Ayrıca, logit modelin esnek modellere göre düşük örnekleme düzeylerinde
daha başarısız tahminler yaptığı görülmüştür.
FULL TEXT (PDF):
- 6
61-73