PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING ARTIFICIAL NEURAL NETWORK

Abstract

The objective of this research was to investigate the usefulness of artifi cial neural network (ANN) in the prediction of slaughter value of young crossbred bulls based on growth data. The studies were carried out on 104 bulls fattened from 120 days of life until the weight of 500 kg. The bulls were group fed using mainly farm feeds. After slaughter the carcasses were dissected and meat was subjected to physico-chemical and organoleptic analyses. The obtained data were used for the development of an artifi cial neural network model of slaughter value prediction. It was found that some slaughter value traits (hot carcass, cold half-carcass, neck and round weights, bone content in dissected elements in half-carcass, meat pH, dry-matter and protein contents in meat and meat tenderness and juiciness) can be predicted with a considerably high accuracy using the artifi cial neural network.

Full text

ORIGINAL PAPER
133
Volume 6 (2005) No. 2 (133-142)
PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING
ARTIFICIAL NEURAL NETWORK
PRZEWIDYWANIE WARTOŚCI RZEŹNEJ BUHAJKÓW Z WYKORZYSTANIEM
SZTUCZNYCH SIECI NEURONOWYCH
Krzysztof ADAMCZYK1*, Krzysztof MOLENDA2, JAN SZAREK1, Grzegorz SKRZYŃSKI1
1Department of Cattle Breeding, Agricultural University, al. Mickiewicza 24/28, 30-059 Kraków, Poland
*Corresponding author. Tel.: ++48 12 622 40 90, Fax: ++48 12 662 41 62, e-mail: rzadamcz@cyf-kr.edu.pl
2Department of Agricultural Mechanization, Agricultural University, ul. Balicka 104, 30-149 Kraków, Poland
Manuscript received: March 19, 2005; Reviewed: April 4, 2005; Accepted for publication: April 27, 2005
ABSTRACT
The objective of this research was to investigate the usefulness of artifi cial neural network (ANN) in the prediction
of slaughter value of young crossbred bulls based on growth data. The studies were carried out on 104 bulls fattened
from 120 days of life until the weight of 500 kg. The bulls were group fed using mainly farm feeds. After slaughter the
carcasses were dissected and meat was subjected to physico-chemical and organoleptic analyses. The obtained data
were used for the development of an artifi cial neural network model of slaughter value prediction. It was found that
some slaughter value traits (hot carcass, cold half-carcass, neck and round weights, bone content in dissected elements
in half-carcass, meat pH, dry-matter and protein contents in meat and meat tenderness and juiciness) can be predicted
with a considerably high accuracy using the artifi cial neural network.
KEYWORDS: bulls; beef performance; neural networks
STRESZCZENIE
W pracy przedstawiono możliwości, jakie daje zastosowanie sztucznych sieci neuronowych do określania wartości
rzeźnej buhajków mieszańców na podstawie parametrów ich wzrostu. Badaniami objęto 104 buhajki mieszańce, które
opasano od 120 dnia życia do uzyskania przez nie ok. 500 kg netto. W tym czasie obowiązywał ujednolicony system
żywienia grupowego w oparciu o pasze gospodarskie. Następnie zwierzęta były ubijane, ich tusze dysekowane a mięso
poddawane analizom fi zyko-chemicznym i organoleptycznym. Przy użyciu sztucznej sieci neuronowej określono
cechy wartości rzeźnej na podstawie parametrów wzrostu buhajków. Wykazano, że szereg ważnych cech wartości
rzeźnej (m.in. masa półtuszy zimnej, karkówki, mięsa udźca, udział kości w półtuszy oraz pH mięsa) mogą być
określane ze stosunkowo dużą dokładnością poprzez zastosowanie sztucznej sieci neuronowej.
SŁOWA KLUCZOWE: buhajki, użytkowość mięsna, sztuczne sieci neuronowe

134 Journal of Central European Agriculture Vol 6 (2005) No 2
Krzysztof ADAMCZYK, Krzysztof MOLENDA, JAN SZAREK, Grzegorz SKRZYŃSKI
INTRODUCTION
Artifi cial neural networks (ANNs) are a calculation
technique originated from the structure and function
of a human brain. The basic propriety of ANNs is a
possibility of less or more self-reliant investigation of the
relationships between sets of inputs and corresponding
sets of outputs through the determination of functional
dependencies with any degree of complexity. It takes
place in, so called, network self-learning process which
is a gradual generalization of the information on a given
phenomenon along with the investigation of particular
real-time relationships among variables ([2]; [6]; [17]).
ANNs perform particularly well in the detection and
incorporation of non-linear relationships and can be
applied to a wide variety of fi elds ([8]).
Many authors found a high effectiveness of ANNs
application in cattle breeding. They have been used for
mastitis prediction ([12]; [25]; [26]), milk, fat and protein
yield prediction ([7]; [8]; [16]; [20]), estimation of somatic
cell count and fat and protein content in milk ([24]),
evaluation of a physiological status of cows (oestrus,
calving and health status) ([11]; [19]) and analysis of in
vitro embryo development ([23]). Neural network models
were also developed for predicting and determination
of an objective measurement of slaughter value in beef
cattle using pre-slaughter information ([3]; [4]; [5]; [9]).
Also, the results of the preliminary investigations carried
out by Adamczyk (2002) indicated a high effectiveness
of ANN in the evaluation of a slaughter value of young
bulls ([1]).
The aim of the presented study was to evaluate the
feasibility of predicting the slaughter value of young
crossbred bulls with ANN using the data on the course
of their growth.
MATERIAL AND METHODS
The investigations were carried out on 104 young bulls
representing the following genetic groups: Black-and-
White × Piemontese (11 bulls), Black-and-White ×
Hereford (11), Black-and-White × Limousine (28),
Black-and-White × Aberdeen Angus (11), Black-and-
White × Charolaise (10), Red-and-White × Charolaise
(11), Red-and-White × Limousine (9), Red-and-White
× Red Angus (8) and (Black-and-White, Polish Red,
Simmental) × Salers (5).
Fattening of bulls and evaluation of their slaughter value
were carried out according to the method employed in
breeding value estimation at beef bulls stations in Poland
([18]). The bulls stayed at the farms where they had been
born up to the age of 2-4 months. The controlled fattening
from the age of 120 days was carried out at the station.
The bulls were fattened until achieved 500 kg of weight,
that means up to 16-18 month of life. They were kept
in stanchion barns, with straw bedding and have free
access to water. The same system of feeding was applied
in every genetic group. The bulls were fed mainly with
farm feeds supplemented with a concentrate. The share
of particular feeds in total dry-matter of feed ration was
following: grass silage -35% , hay -35% and concentrate
-30%. The weighing of bulls were done at birth, at the
day of purchase and then every each month.
At the end of fattening the bulls were slaughtered and
the carcasses were subjected to the dissection according
to the method used in meat industry ([13]; [14]; [15]).
Seven days after slaughter the physico-chemical and
organoleptic analyses were performed with the methods
conventionally used in meat quality evaluation ([ [21];
[22]; [27]).
The ANN used in the presented study was characterized by
the following parameters: feed-forward, back-propagation
training algorithm, one hidden layer (30 hidden neurons),
random choice of starting values of weights ranged from
–1 to 1, constant learning coeffi cient of: 0.2, logistic
function of neuron activation, 10 000 training cycles. The
choice of the ANN type was done based on the results of
preliminary investigations ([1]).
The input layer of the model consisted of the nodes
corresponding to the following variables: genetic group,
age and body weights of a bull - in fattening period and
at slaughter. The output layer (representing the variables
that are being predicted) consisted of the nodes related
to the following slaughter value traits: weights of hot
carcass, cold half-carcass, neck, brisket, fl ank, best ribs,
shoulder meat, fore-ribs, sirloine (T-bone) meat1, fi llet,
round meat, 2nd class meat, bone content in the elements
dissected from a half-carcass (BCED)2, fat content in the
elements dissected from a half-carcass (FCED)3, meat
pH, meat water-holding capacity, meat colour brilliance,
contents of dry matter, fat and protein in meat, marbling,
tenderness and juiciness of meat.
Data set was separated at random into training and
testing data sets (the former was used to train the ANN,
1 – m. long. dorsi extracted from sirloine (T-bone)
2 – elements dissected from a half-carcass: neck, shoulder, sirloine (T-bone), round and hindshin
3 – subcutaneous and intramuscular fat in the elements dissected from a half-carcass

PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING ARTIFICIAL NEURAL NETWORK
135
J. Cent. Eur. Agric. (2005) 6:2, 133-142
and the latter to validate it). Sixty percent (934 records)
were allocated to training, and 40% (620 records) to
validation.
Because the logistic function of neuron activation in the
hidden layer was chosen, the trait values were normalized
between 0 and 1 prior to use with the model, according to
the following formula ([10]):
minmax
min
tt
)tt(
)t(xx 


where: t – original value of a trait, x – normalized value,
tmax, tmin – maximum and minimum values of a trait, both
for training and testing sets.
The fi nal step in network activity was the denormalization
of outputs through their multiplication by (tmax – tmin) and
addition of tmin.
The accuracy of ANN predictions was evaluated using
the coeffi cients of linear correlation and calculated of
the differences between the actual values of traits and
the corresponding ANN predictions. The classes of
distribution of prediction differences together with the
related percentage of predicted values were determined
for each trait.
RESULTS AND DISCUSSION
The estimates of the differences and correlation
coeffi cients between the actual values of slaughter traits
and ANN predictions were the following (Tab. 1-4,
Fig.1):
• for hot carcass weight: the total prediction difference
varied between –12.2 and +13.0 kg and r=0.97. The
particular values of differences were relatively regularly
distributed within the set of prediction differences,
therefore, the determination of more numerous class of
predictions was diffi cult.
• for cold half-carcass: the total prediction difference
varied between –17.6 and +14.2 kg (r=0.92) but most
predicted values (87.4%) ranged from –7.9 to + 7.9 kg in
relation to the actual values in the testing set (AV);
• for neck weight: the total prediction difference varied
between –1.65 and +1.50 kg (r=0.88) and 84.6% of
the predicted values ranged from –1.01 to +0.87 kg in
relation to AV;
• for brisket weight: the total prediction difference varied
between –1.32 and +2.44 kg (r=0.05) and 79.1% of
the predicted values ranged from –0.56 to +0.94 kg in
relation to AV;
• for fl ank weight: the total prediction difference varied
between –2.65 and +3.13 kg (r=0.01) and 80,0% of
the predicted values ranged from –1,48 to +1,40 kg in
relation to AV;
• for best ribs weight: the total prediction difference
varied between –1.12 to +0.98 kg (r=0.37) and 85.1%
of the predicted values ranged from –0.69 to +0.77 kg in
relation to AV;
• for shoulder meat weight: the total prediction difference
varied between –1.31 and +0.96 kg (r=0.49) and 89.9%
of the predicted values ranged from –1.07 to +0.74 kg in
relation to AV;
• for fore-ribs weight: the total prediction difference
varied between –2.50 and +1.45 kg (r=0.60) and 88.1%
of the predicted values ranged from –0.91 to +1.06 kg in
relation to AV;
• for sirloin (T-bone) meat weight: the total prediction
difference varied between –0.97 and +1.21 kg (r=0.61)
and 87.5% of the predicted values ranged from –0.53 to
+0.55 kg in relation to AV;
• for fi llet weight: the total prediction difference varied
between –0.31 and +0.30 kg (r=0.60) and 83.0% of
the predicted values ranged from –0.18 to +0.12 kg in
relation to AV;
• for round meat weight: the total prediction difference
varied between –3.49 and +3.18 kg (r=0.81) and 93.2%
of the predicted values ranged from –2.15 to +1.84 kg in
relation to AV;
• for 2nd class meat weight: the total prediction difference
varied between –9.45 and +15.50 kg (r=0.82) and 89.7%
of the predicted values ranged from –4.45 to +5.52 kg in
relation to AV;
• for BCED: the total prediction difference varied
between –2.08 and +1.84% (r=0.67) and most 90.7%
of the predicted values ranged from –0.90 to +1.05% in
relation to AV;
• for FCED: the total prediction difference varied
between –1.35 and +1.74% (r=0.86) and most 92.3%
of the predicted values ranged from –1.03 to +1.12% in
relation to AV;
• for meat pH: the total prediction difference varied
between –0.9 and +0.6 (r=0.91) and most 86.7% of the
predicted values ranged from –0.4 to +0.3 in relation to
AV;
• for meat water-holding capacity: the total prediction
difference varied between –3.57 and +3.69 cm2 (r=0.81)
and 92.8% of the predicted values ranged from –2.11 to
+2.24 cm2 in relation to AV;
• for meat colour brilliance: the total prediction difference
varied between –3.1 and +4.2% (r=0.87) and 80.1% of the
predicted values ranged from –1.6 to +2.0% in relation to
AV;
• for dry-matter content in meat: the total prediction

136 Journal of Central European Agriculture Vol 6 (2005) No 2
Krzysztof ADAMCZYK, Krzysztof MOLENDA, JAN SZAREK, Grzegorz SKRZYŃSKI
Trait
Mean actual
value in the
testing set
Mean
difference
Standard
deviation of
difference
Minimum
difference
Maximum
difference
Hot carcass weight (kg) 273.6 -0.70 6.67 -12.2 13.0
Cold half-carcass weight (kg) 135.0 0.60 5.45 -17.6 14.2
Neck weight (kg) 9.61 -0.06 0.68 -1.65 1.50
Brisket weight (kg) 7.19 0.02 0.62 -1.32 2.44
Flank weight (kg) 11.57 -0.15 1.16 -2.65 3.13
Best ribs weight (kg) 5.42 -0.04 0.51 -1.12 0.98
Shoulder meat weight (kg) 6.90 -0.05 0.60 -1.31 0.96
Fore-ribs weight (kg) 7.86 -0.02 0.71 -2.50 1.45
Sirloine (T-bone) meat weight (kg) 3.53 0.03 0.38 -0.97 1.21
Fillet weight (kg) 1.52 -0.05 0.12 -0.31 0.30
Round meat weight (kg) 18.93 -0.20 1.24 -3.49 3.18
2nd class meat weight (kg) 39.62 0.08 3.51 -9.45 15.50
BCED (%) 12.18 0.20 0.62 -2.08 1.84
FCED (%) 3.32 -0.02 0.67 -1.35 1.74
Meat pH 6.01 -0.10 0.25 -0.90 0.60
Meat water-holding capacity (cm2) 7.23 0.19 1.34 -3.57 3.69
Meat colour brilliance (%) 12.9 0.40 1.47 -3.1 4.2
Dry-matter content in meat (%) 23.91 -0.09 0.74 -1.90 2.30
Fat content in meat (%) 1.47 0.13 0.64 -2.43 3.03
Protein content in meat (%) 21.24 -0.02 0.58 -1.51 2.24
Meat marbling (point) 2.0 -0.10 0.39 -1.0 2.2
Meat tenderness (point) 4.5 -0.10 0.28 -1.0 1.0
Meat juiciness (point) 4.6 -0.10 0.22 -0.7 0.8
Table 1. Differences between the actual values of slaughter traits and predicted by ANN
difference varied between –1.90 and +2.30% (r=0.81)
and 88.5% of the predicted values ranged from –1.05 to
+1.04 kg in relation to AV;
• for fat content in meat: the total prediction difference
varied between –2.43 and +3.03% (r=0.83) and 87.1%
of the predicted values ranged from –0.78 to +0.85% in
relation to AV;
• for protein content in meat: the total prediction
difference varied between –1.51 and +2.24% (r=0.83)
and 85.9% of the predicted values ranged from –0.75 to
+0.74% in relation to AV;
• for meat marbling : the total prediction difference
varied between –1.0 and +2.2 points (r=0.88) and 86.4%
of the predicted values ranged from –0.3 to +0.6 points
in relation to AV;
• for meat tenderness: the total prediction difference
varied between –1.0 and +1.0 points (r=0.90) and 85.0%
of the predicted values ranged from –0.4 to +0.4 kg in
relation to AV;
• for meat juiciness: the total prediction difference varied
between –0.7 and +0.8 points (r=0.91) and 90.7% of the
predicted values ranged from –0.4 to +0.3 kg in relation
to AV.
The presented results indicate that the prediction ability
of ANN, characterized by the magnitude of correlation
coeffi cient and the difference between the actual and the
predicted value of a trait, was the highest for hot carcass
weight and only slightly lower for many other important
traits of slaughter value, such as: cold half-carcass weight,
meat pH, juiciness and tenderness.
The predicted values for neck weight, round meat
weight, 2nd class meat weight, FCED, meat water holding
capacity, meat colour brilliance, dry-matter, fat and
protein contents in meat and meat marbling were highly
correlated with the actual ones (r=0.81-0.89) however,
the prediction differences estimated for those traits
proved to be considerable. Within this group the highest
accuracies of ANN predictions were found for dry-matter
and protein contents in meat.
The effi ciency of ANN in the prediction of best ribs and
shoulder meat weights was very poor. The prediction of
fl ank and brisket weights appeared to be impossible but
those cuts are of low culinary value.
In summary, it can be said that the differences in the ability

PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING ARTIFICIAL NEURAL NETWORK
137
J. Cent. Eur. Agric. (2005) 6:2, 133-142
Traits
Hot carcass weight Cold half-carcass weight Neck weight Brisket weight
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of prediction
differences
(kg)
Number of
predicted
values
(%)
[-12.3 ; -9.7] 9.8 [-17.7 ; -14.4] 1.6 [-1.75 ; -1.33] 3.1 [-1.42 ; -0.95] 5.3
[-9.6 ; -7.1] 9.7 [-14.3 ; -11.2] 1.0 [-1.32 ; -1.02] 4.5 [-0.94 ; -0.57] 9.8
[-7.0 ; -4.6] 8.9 [-11.1 ; -8.0] 1.8 [-1.01 ; -0.70] 11.3 [-0.56 ; -0.19] 26.8
[-4.5 ; -2.1] 16.6 [-7.9 ; -4.9] 11.1 [-0.69 ; -0.39] 14.8 [-0.18 ; 0.18] 16.3
[-2.0 ; 0.4] 12.7 [-4.8 ; -1.7] 17.9 [-0.38 ; -0.07] 15.0 [0.19 ; 0.56] 22.1
[0.5 ; 2.9] 16.0 [-1.6 ; 1.5] 19.7 [-0.06 ; 0.24] 13.5 [0.57 ; 0.94] 13.9
[3.0 ; 5.5] 6.8 [1.6 ; 4.7] 25.5 [0.25 ; 0.56] 16.1 [0.95 ; 1.31] 3.7
[5.6 ; 8.0] 6.0 [4.8 ; 7.9] 13.2 [0.57 ; 0.87] 13.9 [1.32 ; 1.69] 1.1
[8.1 ; 10.5] 5.8 [8.0 ; 11.1] 6.6 [0.88 ; 1.19] 5.6 [1.70 ; 2.07] 0.0
[10.6 ; 13.1] 7.7 [11.2 ; 14.3] 1.6 [1.20 ; 1.51] 2.1 [2.08 ; 2.45] 1.0
Traits
Flank weight Best ribs weight Shoulder meat weight Fore-ribs weight
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of
prediction
differences (kg)
Number of
predicted
values
(%)
Classes of prediction
differences
(kg)
Number of
predicted
values
(%)
[-2.75 ; -2.07] 3.2 [-1.22 ; -0.91] 4.0 [-1.41 ; -1.08] 3.2 [-2.60 ; -2.10] 1.0
[-2.06 ; -1.49] 8.2 [-0.90 ; -0.70] 6.0 [-1.07 ; -0.85] 9.8 [-2.09 ; -1.71] 0.0
[-1.48 ; -0.92] 17.6 [-0.69 ; -0.49] 13.2 [-0.84 ; -0.63] 12.1 [-1.70 ; -1.31] 1.3
[-0.91 ; -0.34] 18.1 [-0.48 ; -0.28] 13.7 [-0.62 ; -0.40] 9.2 [-1.30 ; -0.92] 5.5
[-0.33 ; 0.24] 14.5 [-0.27 ; -0.07] 11.6 [-0.39 ; -0.17] 4.2 [-0.91 ; -0.52] 21.1
[0.25 ; 0.82] 13.2 [-0.06 ; 0.14] 10.3 [-0.16 ; 0.05] 9.7 [-0.51 ; -0.13] 13.9
[0.83 ; 1.40] 16.6 [0.15 ; 0.35] 16.6 [0.06 ; 0.28] 16.8 [-0.12 ; 0.27] 16.6
[1.41 ; 1.97] 6.0 [0.36 ; 0.56] 9.7 [0.29 ; 0.51] 12.3 [0.28 ; 0.66] 26.3
[1.98 ; 2.55] 0.8 [0.57 ; 0.77] 10.0 [0.52 ; 0.74] 15.8 [0.67 ; 1.06] 10.2
[2.56 ; 3.14] 1.8 [0.78 ; 0.99] 4.8 [0.75 ; 0.97] 6.9 [1.07 ; 1.46] 4.2
Table 2. Distribution of differences between the actual values of slaughter traits and predicted by ANN
of ANN to predict the analysed traits was considerable.
However, the obtained results should be treated as
one of the preliminary attempts of ANN application
for the prediction of bulls’ slaughter value using the
growth data. For ANN to fulfi ll its potential in this area,
continued efforts must be made. The improvement of
ANN prediction ability could be achieved through: the
elimination or addition of input and output variables, data
preprocessing, increase of the number of hidden neurons
and the number of their layers, increase of the number of
ANN training cycles and change of the method of ANN
training process.
The possibility of an effi cient application of ANNs for the
prediction of beef slaughter value was also investigated
by other authors. Brethour (1994) found similar results of
marbling score estimation in live bulls from ultrasound
images using pattern recognition and Neural Network
procedures ([3]). The comparison of sensory evaluation of
meat tenderness after slaughter with the evaluation based
on ultrasonic images of colour brilliance, marbling and
structure of meat made by Li et al. (1999) also proved the
effi ciency of the models generated by means of Neural
Networks in the interpretation of ultrasonic images ([9]).
When evaluating the degree of cartilage ossifi cation in
the thoracic vertebrae, that could be used as a predictor
of a slaughter value, Hatem and Tan (1998) successfully
used ANNs in the interpretation of the relative vertebrae
images ([4]). Hill et al. (2000) developed Neural Network
models for predicting and classifying an objective
measurement of meat tenderness using considerably
numerous information including: sex, slaughter weight,
hot carcass weight, meat colour brilliance, area of
musculus longissimus dorsi cross-section, marbling and
meat cooking method ([5]).
CONCLUSIONS
1. There is a possibility of the effi cient prediction
of hot carcass weight of young beef bulls from growth
data using artifi cial neural network.

138 Journal of Central European Agriculture Vol 6 (2005) No 2
Krzysztof ADAMCZYK, Krzysztof MOLENDA, JAN SZAREK, Grzegorz SKRZYŃSKI
Table 3. Distribution of differences between the actual values of slaughter traits and predicted by ANN
Traits
Sirloine (T-bone) meat weight Fillet weight Round meat weight 2nd class meat weight
Classes of
prediction
differences
(kg)
Number of predicted
values
(%)
Classes of
prediction
differences
(kg)
Number of
predicted
values
(%)
Classes of prediction
differences (kg)
Number of
predicted
values
(%)
Classes of
prediction
differences
(kg)
Number of
predicted
values
(%)
[-1.07 ; -0.76] 1.8 [-0.41 ; -0.25] 2.3 [-3.59 ; -2.83] 1.9 [-9.55 ; -6.95] 3.7
[-0.75 ; -0.54] 4.4 [-0.24 ; -0.19] 6.8 [-2.82 ; -2.16] 0.5 [-6.94 ; -4.46] 4.0
[-0.53 ; -0.32] 15.3 [-0.18 ; -0.13] 21.6 [-2.15 ; -1.49] 9.2 [-4.45 ; -1.96] 18.4
[-0.31 ; -0.10] 12.3 [-0.12 ; -0.07] 18.4 [-1.48 ; -0.82] 24.5 [-1.95 ; 0.53] 25.3
[-0.09 ; 0.12] 22.3 [-0.06 ; -0.01] 13.2 [-0.81 ; -0.16] 19.0 [0.54 ; 3.03] 34.7
[0.13 ; 0.33] 21.0 [0.0 ; 0.06] 10.6 [-0.15 ; 0.51] 13.2 [3.04 ; 5.52] 11.3
[0.34 ; 0.55] 16.6 [0.07 ; 0.12] 19.2 [0.52 ; 1.18] 19.7 [5.53 ; 8.02] 0.8
[0.56 ; 0.77] 5.3 [0.13 ; 0.18] 5.6 [1.19 ; 1.84] 7.6 [8.03 ; 10.51] 0.8
[0.78 ; 0.99] 0.0 [0.19 ; 0.24] 1.5 [1.85 ; 2.51] 1.5 [10.52 ; 13.00] 0.0
[1.00 ; 1.22] 1.1 [0.25 ; 0.31] 0.8 [2.52 ; 3.19] 2.9 [13.01 ; 15.51] 1.0
Traits
BCED FCED Meat pH Meat water-holding capacity
Classes of
prediction
differences
(%)
Number of
predicted values
(%)
Classes of
prediction
differences
(%)
Number of
predicted
values
(%)
Classes of
prediction
differences
Number of
predicted
values
(%)
Classes of
prediction
differences
(cm2)
Number of
predicted
values
(%)
[-2.18 ; -1.69] 1.0 [-1.45 ; -1.04] 3.4 [-0.98 ; -0.73] 1.1 [-3.67 ; -2.84] 1.0
[-1.68 ; -1.30] 0.0 [-1.03 ; -0.73] 11.6 [-0.72 ; -0.58] 1.8 [-2.83 ; -2.12] 1.3
[-1.29 ; -0.91] 1.5 [-0.72 ; -0.42] 20.0 [-0.57 ; -0.43] 4.8 [-2.11 ; -1.39] 11.5
[-0.90 ; -0.52] 8.2 [-0.41 ; -0.11] 15.2 [-0.42 ; -0.29] 7.1 [-1.38 ; -0.66] 19.5
[-0.51 ; -0.12] 16.1 [-0.10 ; 0.20] 10.3 [-0.28 ; -0.14] 15.2 [-0.65 ; 0.06] 13.2
[-0.11 ; 0.27] 32.4 [0.21 ; 0.51] 13.4 [-0.13 ; 0.01] 24.8 [0.07 ; 0.79] 14.4
[0.28 ; 0.66] 15.3 [0.52 ; 0.81] 13.4 [0.02 ; 0.16] 26.1 [0.80 ; 1.51] 19.8
[0.67 ; 1.05] 18.7 [0.82 ; 1.12] 8.4 [0.17 ; 0.30] 13.5 [1.52 ; 2.24] 14.4
[1.06 ; 1.44] 3.2 [1.13 ; 1.43] 3.2 [0.31 ; 0.45] 3.1 [2.25 ; 2.97] 4.5
[1.45 ; 1.85] 3.5 [1.44 ; 1.75] 1.1 [0.46 ; 0.70] 2.4 [2.98 ; 3.70] 0.5
2. Some other important traits of slaughter value
(cold half-carcass, neck and round meat weights, bone
content in dissected elements in half-carcass, meat
pH, dry matter and protein contents in meat and meat
tenderness and juiciness) can also be predicted with a
relatively high accuracy.
3. The artifi cial neural network can be treated as
an interesting alternative to traditional models when
examining animal production data.
4. The obtained results are encouraging but future
research on optimization of neural confi guration, choice
of input and outputs and possible data preprocessing is
necessary to increase the prediction accuracy.
REFERENCES
[1] Adamczyk K. Określanie masy ciała, tempa
wzrostu i wartości rzeźnej buhajków mieszańców. Ph.D.
Thesis, Dep. of Cattle Breeding, Agricultural University
in Krakow. 2002.
[2] Azam F. Biologically Inspired Modular Neural
Networks. Ph.D. Thesis, Blacksburg, Virginia, http://
scholar.lib.vt.edu/theses/available/etd-06092000-
12150028/unrestricted/etd.pdf . 2000.
[3] Brethour J.R. Estimating marbling score in live
cattle from ultrasound images using pattern recognition
and neural network procedures. J. Anim. Sci. 72 (6),
(1994), 1425-1432.
[4] Hatem I., Tan J.L. Determination of animal
skeletal maturity by image processing. In: ASAE Annual
International Meeting, Orlando, Florida, 12-16 July,
ASAE Paper, nr 983019. (1998), pp. 13.
[5] Hill B.D., Jones S.D.M., Robertson W.M., Major
I.T. Neural network modeling of carcass measurements
to predict beef tenderness. Can. J. Anim. Sci. 80, (2000),
311-318.
[6] Klerfors D. Artifi cial neural networks: What

PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING ARTIFICIAL NEURAL NETWORK
139
J. Cent. Eur. Agric. (2005) 6:2, 133-142
Table 4. Distribution of differences between the actual values of slaughter traits and predicted by ANN
Traits
Meat colour brilliance Dry-matter content in meat Fat content in meat Protein content in meat
Classes of
prediction
differences
(%)
Number of
predicted
values
(%)
Classes of
prediction
differences
(%)
Number of
predicted
values
(%)
Classes of
prediction
differences
(%)
Number of
predicted
values
(%)
Classes of
prediction
differences
(%)
Number
of
predicted
values
(%)
[-3.2 ; -2.4] 1.5 [-2.00 ; -1.48] 1.5 [-2.53 ; -1.88] 0.8 [-1.61 ; -1.13] 1.0
[-2.3 ; -1.7] 6.6 [-1.47 ; -1.06] 4.8 [-1.87 ; -1.34] 0.2 [-1.12 ; -0.76] 5.3
[-1.6 ; -1.0] 14.8 [-1.05 ; -0.64] 21.9 [-1.33 ; -0.79] 0.6 [-0.75 ; -0.38] 23.9
[-0.9 ; -0.2] 14.0 [-0.63 ; -0.22] 20.3 [-0.78 ; -0.25] 27.9 [-0.37 ; -0.01] 21.0
[-0.1 ; 0.5] 16.3 [-0.21 ; 0.20] 13.5 [-0.24 ; 0.30] 38.9 [0.0 ; 0.37] 25.0
[0.6 ; 1.2] 14.5 [0.21 ; 0.62] 20.2 [0.31 ; 0.85] 20.3 [0.38 ; 0.74] 16.0
[1.3 ; 2.0] 20.5 [0.63 ; 1.04] 12.6 [0.86 ; 1.39] 8.4 [0.75 ; 1.12] 5.3
[2.1 ; 2.7] 6.9 [1.05 ; 1.46] 3.2 [1.40 ; 1.94] 1.3 [1.13 ; 1.49] 0.0
[2.8 ; 3.4] 1.9 [1.47 ; 1.88] 0.3 [1.95 ; 2.48] 0.3 [1.50 ; 1.87] 1.6
[3.5 ; 4.3] 2.9 [1.89 ; 2.31] 1.6 [2.49 ; 3.04] 1.3 [1.88 ; 2.25] 1.0
Traits
Meat marbling Meat tenderness Meat juiciness
Classes of
prediction differences
(point)
Number of
predicted
values
(%)
Classes of
prediction
differences
(point)
Number of
predicted
values
(%)
Classes of prediction
differences
(point)
Number of
predicted
values
(%)
[-1.1 ; -0.7] 2.3 [-1.07 ; -0.78] 1.0 [-0.79 ; -0.55] 1.9
[-0.6 ; -0.4] 6.6 [-0.77 ; -0.58] 2.4 [-0.54 ; -0.40] 1.9
[-0.3 ; 0.0] 25.6 [-0.57 ; -0.38] 5.3 [-0.39 ; -0.26] 7.9
[0.1 ; 0.3] 31.0 [-0.37 ; -0.18] 17.4 [-0.25 ; -0.11] 27.9
[0.4 ; 0.6] 29.8 [-0.17 ; 0.01] 50.6 [-0.10 ; 0.03] 36.5
[0.7 ; 0.9] 3.1 [0.02 ; 0.21] 8.9 [0.04 ; 0.18] 9.2
[1.0 ; 1.2] 0.3 [0.22 ; 0.41] 8.1 [0.19 ; 0.32] 9.2
[1.3 ; 1.5] 0.3 [0.42 ; 0.61] 4.8 [0.33 ; 0.47] 4.0
[1.6 ; 1.8] 0.0 [0.62 ; 0.80] 0.0 [0.48 ; 0.61] 0.0
[1.9 ; 2.3] 1.0 [0.81 ; 1.10] 1.5 [0.62 ; 0.86] 1.5
are they? How do they work? In what areas are they
used?, http://hem.hj.se/~de96klda/NeuralNetworks.htm.,
(1998).
[7] Lacroix R., Salehi F., Yang X.Z., Wade K.M.
Effects of data preprocessing on the performance of
artifi cial neural networks for dairy yield prediction and
cow culling classifi cation. Transactions of the ASAE, 42
(4), (1997), 839-846.
[8] Lacroix R., Wade K.M., Kok R., Hayes J.F.
Prediction of cow performance with a connectionist
model. Transactions of the ASAE, 42 (4), (1995), 1573-
1579.
[9] Li J., Tan J., Martz F.A., Heymann H. Image
texture features as indicators of beef tenderness. Meat
Sci., 53 (1), (1999), 17-22.
[10] Masters T. Sieci neuronowe w praktyce –
Programowanie w języku C++ (Practical Neural Network
Recipes in C++), Wyd. Nauk. Techn., Warszawa,
(1993).
[11] Molenda K., Gil Z., Żychlińska J., Conception
of computer system support in detection of pregnancy,
oestrus and mastitis based on milk temperature. In: 52nd
Annual Meeting of EAAP, Budapest, Hungary, 26-29
August, (2001), Wageningen Pers, 132.
[12] Nielen M., Spigt M.H., Schukken Y.H., Deluyker
H.A., Maatje K., Brand A. Application of a Neural
Network to analyse on-line milking parlour data for the
detection of clinical mastitis in dairy cows. Prev. Vet.
Med. 22, 1-2, (1995), 15-28.
[13] PN-88 A-82003-1. Polska Norma „Wołowina

140 Journal of Central European Agriculture Vol 6 (2005) No 2
Krzysztof ADAMCZYK, Krzysztof MOLENDA, JAN SZAREK, Grzegorz SKRZYŃSKI
0 0,2 0,4 0,6 0,8 1
Correlation coefficient
Hot carcass weight (kg)
Co ld half-c a r c a s s w e ight (kg)
Meat pH
Me a t juic iness (point)
Mea t tenderne ss (point)
Neck weight (kg)
Meat marbling (point)
Me a t c olour brillia nce (%)
Fat content in half-carcass (%)
Protein content in meat (%)
Fat content in meat (%)
2nd class meat we ight (kg)
Round meat weight (kg)
Dry-matter content in meat (%)
Meat water-holding capacity (square cm)
Bone content in half-carcass (%)
Sirloine (T-bone) meat weight (kg)
Fore - r ibs w e ight (kg)
Fillet we ight (kg)
Shoulder meat weight (kg)
Best ribs weight (kg)
Briske t weight (kg)
Flank weight (kg)
0.97
0.92
0.91
0.91
0.90
0.88
0.88
0.87
0.86
0.83
0.83
0.82
0.81
0.81
0.81
0.68
0.67
0.60
0.60
0.49
0.37
0.05
0.01
Fig. 1. Coeffi cients of correlation between the actual values of slaughter traits and predicted by ANN
– części zasadnicze”. (1989).
[14] PN-91 A-82001. Polska Norma „Mięso w
tuszach, półtuszach i ćwierćtuszach”. (1991).
[15] PN-A-82014. Polska Norma „Mięso bez kości do
produkcji przetworów z mięsa rozdrobnionego”. (1997).
[16] Salehi F., Lacroix R., Wade K.M. Improving dairy
yield predictions through combined record classifi ers and
specialized artifi cial neural networks. Comput. Electron.
Agric. 20 (3), (1998), 199-213.
[17] Tadeusiewicz R. Sieci neuronowe. Akademicka
Ofi cyna Wydawnicza, Warszawa. (1993), pp. 195.
[18] Trela J., Choroszy Z., Wajda S., Puchajda Z.,
Metodyka – Ocena wartości hodowlanej buhajów ras
mięsnych metodą stacjonarną na podstawie użytkowości
potomstwa. IZ Balice, (1996), typewriting, pp. 3.
[19] Vassileva S.T., Radev D. Application of

PREDICTION OF BULLS’ SLAUGHTER VALUE FROM GROWTH DATA USING ARTIFICIAL NEURAL NETWORK
141
J. Cent. Eur. Agric. (2005) 6:2, 133-142
neural networks in dairy husbandry. Biotech. in Anim.
Husbandry. 17, 5-6, (2001), 287-294.
[20] Wade K.M., Lajoie L., Lafontaine S., Doyle P.
The role of artifi cial intelligence in decision making based
on farm records. In: 29th Biennale Session of ICAR,
Ottawa, 31 July-5 August, 1994, (1995), 329-333.
[21] Wajda S. Ocena i wykorzystanie surowców
pochodzenia zwierzęcego. Przewodnik do ćwiczeń.
Skrypt AR-T w Olsztynie, (1986). pp. 117.
[22] Wichłacz H. Skup i ocena bydła rzeźnego. Wyd.
Centrum Doradztwa i Edukacji w Rolnictwie, Poznań.
(1996), pp. 71.
[23] Wilkinson R.F., Ming R., Anderson B., Bunch
T.D., White K.L. The use of Neural Networks in
developing novel embryo culture media-formulations.
Theriogenology, 45, 1, (1996), 41-49.
[24] Whyte D., Claycomb R., Kunnemeyer R.
Measurement of somatic cell count, fat and protein in milk
using visible to near infra-red spectroscopy. In: ASAE
Annual International Meeting, Milwaukee, Wisconsin,
USA, 9-12 July, (2000), 1-11.
[25] Yang X.Z., Lacroix R., Wade K.M. Neural
detection of mastitis from dairy herd improvement
records. Transactions of the ASAE, 42 (4), (1999), 1063-
1071.
[26] Yang X.Z., Wade K.M., Lacroix R. Investigation
into the production and conformation traits associated
with clinical mastitis using artifi cial neural networks.
Can. J. Anim. Sci. 80, (2000), 415-426.
[27] Znaniecki P. Zarys obrotu, oceny i przetwórstwa
surowców pochodzenia zwierzęcego. PWRiL, Warszawa,
(1983), pp. 414.