Comparison of the Hebbian Algorithm Based on Input and Output
Patterns in the Prediction of Lung Cancer in Smokers
Tantri Hidayati Sinaga
1
, Arie Rafika Dewi
1
, Masdiana Sagala
2
, Yulia Agustina Dalimunthe
3
and Solikhun
4
1
Universitas Harapan Medan, Medan, Indonesia
2
Universitas Katolik Santo Thomas, Medan, Indonesia
3
Politeknik Negeri Medan, Medan, Indonesia
4
STIKOM Tunas Bangsa, Pematangsiantar, Indonesia
Keywords:
Hebbian Algorithm, Input and Output Patterns, Prediction of Lung Cancer.
Abstract:
The Hebbian algorithm learning method is a learning method that is carried out by fixing the weight values
so that if there are 2 neurons connected and both are alive at the same time, the weight between the two is
increased. This study’s main problem is finding the best performance of the Hebbian algorithm to predict
lung cancer in smokers. There are 15 attributes to determine lung cancer, namely: gender(x1), age(x2), smok-
ing(x3), yellow finger(x4), anxiety(x5), social pressure(x6), chronic diseases(x7), fatigue(x8), allergies(x9 ),
wheezing(x10), consumption of alcohol(x11), cough (x12), shortness of Breath(x13), hard to swallow(x14)
and chest Pain(x15). This study compares the Hebbian algorithm with four forms of test simulation with four
states of input and output patterns. The test simulation results show the best accuracy is with binary data input
patterns and binary and bipolar output patterns. The accuracy obtained is 65%.
1 INTRODUCTION
Lung cancer is an abnormal condition found in the
lungs characterized by abnormal cell growth or what
is known as a dangerous tumor. Abnormal cell growth
conditions can originate from cells present in the
lungs. However, this abnormal cell growth can origi-
nate from cancer cells in other body parts that spread
to the lungs. Lung cancer can occur in both men
and women (Bisri et al., 2013), (Prasetio and Susanti,
2019), (Setyadi et al., 2020).
Research (Pedersen and Risi, 2021), the results of
this study add to the argument about generalization,
overfitting, and OOD adaptation. Hebbian learning
mixed with managing ”genomic bottlenecks” could
be a viable research direction for creating agents
adapting to a larger range of unexpected scenarios.
Research (Kwessi, 2022), researchers propose an
inspired synaptic plasticity rule from the Allee ef-
fect, a frequent occurrence in population dynam-
ics. Researchers exhibited properties, namely synap-
tic normalization, weight competition, decorrelation
potential, and satisfactory stability. Researchers have
shown that Allee’s effect on synaptic plasticity can be
enhanced in the lack of plasticity.
Research (Osakabe et al., 2021), the results of the
researcher’s numerical simulation show that proposes
a modified version of the Hebb rule and increases the
anti-Hebb learning achievement. In addition, the re-
searcher revealed that the possibility of taking the tar-
get pattern from several studied patterns is quite high.
Research (Napole et al., 2020), the outcome is ana-
lyzed regarding guidance, mistake, and control sig-
nals, and performance is assessed using the integral
of absolute error (IAE). Experiments reveal that FF-
ANN compensation in conjunction with SNPID is ap-
propriate.
Research (Qin and Duan, 2020), the experimen-
tal finding showed that the single-neuron adaptive
hysteresis compensation approach successfully tracks
continuous and discontinuous trajectories. It outper-
forms the rate-dependence of the PEA’s hysteresis
in adaptive and self-learning performance. Research
(Napitupulu and Situmorang, 2020), redevelop this fi-
nal project by optimizing the application of employee
ownership assessment using the Hebb rule neural net-
Sinaga, T., Dewi, A., Sagala, M., Dalimunthe, Y. and Solikhun, .
Comparison of the Hebbian Algorithm Based on Input and Output Patterns in the Prediction of Lung Cancer in Smokers.
DOI: 10.5220/0012448100003848
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 3rd International Conference on Advanced Information Scientific Development (ICAISD 2023), pages 275-280
ISBN: 978-989-758-678-1
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
275

work approach.
Research (Illing et al., 2021), this research pro-
posed a hybrid fuzzy-rough set strategy called RS-
HeRR to generate effective, editable, and concise rule
sets. It incorporates strict rules fuzzy generation and
reduction systems, called Hebbian-based and novel
rule-based reduction algorithms, attribute selection
algorithms for withdrawal rules. The proposed hy-
bridization uses the pull-through rule consolidation
of partial dependencies and system performance im-
provements to significantly reduce problems redun-
dancy in HeRR. However, it provides similar or better
accuracy. RS-HeRR exhibits these characteristics in
more than four practical classification problems, such
as diabetes assistance, urban air treatment monitoring,
sonar target classification, and ovarian cancer detec-
tion. It showed very good performance-biased data
sets.
Research (Muscoloni and Cannistraci, 2021), the
study came to the conclusion that SPM and a straight-
forward brain bioinspired rule like CH performed
much better than both the AI-created brute-force
method and did not do any better. Stacking is ideal
but incomplete, which is consistent with G
¨
odel’s in-
completeness. What is already in your feature cannot
be pushed any farther. Hench, we also need to work
toward AI that matches a human’s physical make-up
and ”knowledge” of basic, complicated laws. AIs
that ”stole fire from the Gods for mankind,” and are
on their way to machine intelligence, may live in
the future.Research (Liu et al., 2021), this paper pro-
poses a hybrid fuzzy-rough set approach called RS-
HeRR to generate effective, editable, and concise rule
sets. It incorporates strict rules fuzzy generation and
reduction systems, named Hebbian-based and novel
rule-based reduction algorithms, and attribute selec-
tion algorithms for withdrawal rules. The proposed
hybridization uses the pull-through rule consolidation
of partial dependencies and system performance im-
provements to significantly reduce problems redun-
dancy in HeRR. However, it provides similar or bet-
ter accuracy. RS-HeRR exhibits these features in
more than four practical classification issues, namely
diabetes assistance, urban air treatment monitoring,
sonar target classification, and ovarian cancer detec-
tion. It showed a very good product of biased data
sets.
Research (Isomura and Toyoizumi, 2019) demon-
strated that a neural network that implements Heb-
bian error-gated rules (EGHR) with sufficiently re-
dundant sensory input could successfully learn this
task. After training, the network can do multi-context
BSS without further synapse updating by keeping all
memory experienced context. It showed an interest in
EGHR usage for dimension reduction by extracting
low-dimensional resources across contexts. Lastly,
if common features are presented in contexts, EGHR
could identify and categorize them even in green con-
texts. The finding highlighted the usefulness of the
EGHR as a perceptual adaptation in the animal model.
Research (Najarro and Risi, 0 12), researchers
found from a completely random weight, Hebbian
found rules allow agents to navigate dynamic 2D pixel
environments. Furthermore, They enable a simulated
3D quadrupedal robot in less than 100 steps while
compensating for morphological damage not visible
during training and without explicit reward or er-
ror signal. Research (Journ
´
e et al., 2022) SoftHebb
demonstrates with a very different approach from BP
i.e. Deep Learning through multiple layers of sense
in the brain and increasing the accuracy of bio-sense
machine learning.
Research (Tsai, 2021), in the PE-SM system,
which is made of stacked atomic-thick materials,
graphene acts as a charge storage layer, hexagonal
boron nitride as a tunneling dielectric, and rhenium
diselenide as a powerful photosensor. Under the op-
tical fingernail, it executes synaptic metaplasticity
thanks to the PE-SM function. A additional 24 24
PESM are included in the simulated spiking neural
network, which uses Hebbian rules to recognize im-
ages in an unsupervised machine learning environ-
ment. In addition to improving neuromorphic pro-
cessing effectiveness, PE-SM also makes the circuit
size structure more uniform. The electro-photoactive
concept offers a revolutionary method for creating
photonic synaptic devices or other photosensitive 2D
materials, and it can also be used to other photosensi-
tive materialsResearch (Magotra and Kim, 2020) , the
proposed HTL algorithm can increase learning trans-
fer performance, especially in terms of heterogeneous
source and target data, according to experimental re-
sults using the CIFAR-10 (Canadian Institute for Ad-
vanced Research) and CIFAR-100 datasets in various
combinations.
Research (Golkar et al., 0 12) The researchers de-
tailed how, in the researchers’ model, potential cal-
cium plateaus could be interpreted as a signal of back-
propagation error. Researchers demonstrated that, de-
spite relying solely on biologically reasonable local
learning rules, our algorithm performs competitively
with existing implementations of RRMSE and CCA.
Research (Pogodin and Latham, 0 12), the result-
ing rules feature a three-component Hebbian struc-
ture: they call for pre- and post-synaptic firing rates,
and a third factor, an error signal, is made up of a
global teaching signal and a layer-specific term that
are both accessible without top-down access. They
ICAISD 2023 - International Conference on Advanced Information Scientific Development
276

rely on convenience between partners rather than a
suitable label to produce the desired results. Fur-
thermore, our rules demand divided normalization,
a characteristic well-known to tissue biology, in or-
der to achieve strong performance on challenging
challenges while maintaining biological plausibility..
Lastly, the simulations show that our performance
rules are almost the same as well as backpropagation
(Solikhun et al., 2020b), (Solikhun et al., 2020a) in
the image classification task.
Research (Nicola and Clopath, 2019) Researchers
utilized advanced techniques in network iterative
spiking training to demonstrate how especially in-
terneuron networks can: 1) generate internal theta se-
quences for binds external spike elicited in presence
of obstruction from medial septum, 2) compresses
studied spike Sequence in SPW-R form when sep-
tal block is removed, 3) generates and fixes high fre-
quency assembly during SPW-mediated compression,
and 4) timing the interripple between the SPW-Rs in
the ripple cluster. From fast timescales of neurons
to slow timescales of behavior, Network interneurons
serve as a scaffold for one-shot learning by replaying,
inverting, refining, and setting spikes of Sequences.
Research (Gillett et al., 2020), the researcher dis-
covered that non-linearity in the learning rules might
determine how sparsely the recalled sequences were
composed. Additionally, sequences maintain good
decoding while exhibiting very labile dynamics as
synaptic connection is continuously updated as a re-
sult of noise or other pattern storage, comparable to
these recent discoveries in the parietal cortex and hip-
pocamal pus. Finally, the researchers showed that
their findings preserved spiny neuron repeat networks
with distinct excitatory and inhibitory populations.
This study’s main problem is finding the best perfor-
mance of the Hebbian algorithm in predicting lung
cancer in smokers. Researchers signify lung cancer
in smokers by using 15 attributes. Researchers com-
pare the Hebbian algorithm based on input patterns
and output patterns. There are four forms of input
and output patterns, namely binary input and output
patterns, bipolar input and output patterns, binary in-
put and bipolar output patterns and bipolar input and
binary output patterns.
2 RESEARCH METHODOLOGY
2.1 Research Data
Data on lung cancer prediction in smokers is taken
from Kaggle. The data consists of 16 attributes.
Fifteen input data attributes and one target data at-
tribute. The 15 attributes are gender(x1), age(x2),
smoking(x3), yellow finger(x4), anxiety(x5), social
pressure(x6), chronic diseases(x7), fatigue(x8), al-
lergies(x9 ), wheezing(x10), consumption of alco-
hol(x11), cough (x12), shortness of Breath(x13), hard
to swallow(x14) and chest Pain(x15). The target’s at-
tribute is lung cancer. The data used for the simula-
tion test is 20 data. Here is the raw data of lung cancer
prediction in smokers with 20 data.
Table 1: Raw Data on Lung Cancer Prediction in Smokers.
No X1 X2 .. X15 T
1 M 69 .. 1 YES
2 M 74 .. 1 YES
3 F 59 .. 1 NO
4 M 63 .. 1 NO
5 F 63 .. 0 NO
6 F 75 .. 0 YES
7 M 52 .. 1 YES
8 F 50 .. 0 YES
9 F 68 .. 0 NO
10 M 53 .. 1 YES
11 F 60 .. 0 YES
12 M 72 .. 1 YES
13 F 60 .. 0 NO
14 M 58 .. 1 YES
15 M 69 .. 1 NO
16 F 48 .. 0 YES
17 M 75 .. 1 YES
18 M 57 .. 1 YES
19 F 68 .. 0 YES
20 F 60 .. 0 NO
Lung cancer medical record data is converted with
the following rules:
1. Gender:
• If M(Male) then 1;
• If F(Female) then 0
2. Age :
• If old age(46-55 years), late old(56-65 years)
and old age(>65years) then 1; otherwise 0.
3. Smoking:
• If smoking then 1;
• If don’t smoke then 0.
4. Yellow Finger:
• If have jaundice then 1
• If don’t have jaundice then 0.
5. Anxiety:
• If have anxiety disease then 1;
• If don’t have Anxiety disease then 0.
Comparison of the Hebbian Algorithm Based on Input and Output Patterns in the Prediction of Lung Cancer in Smokers
277

6. Peer Pressure :
• If have Peer Pressure then 1;
• If don’t have Peer Pressure then 0.
7. Chronic Disease:
• If have Chronic Disease then 1;
• If don’t have Chronic Disease then 0.
8. Fatigue:
• If have Fatigue then 1;
• If don’t have Fatigue then 0.
9. Allergies:
• If have an allergic disease then 1;
• If don’t have an allergic disease then 0.
10. Wheezing :
• If have Wheezing disease then 1;
• If do not have Wheezing disease then 0.
11. Consumption of alcohol:
• If consuming alcohol then 1;
• If don’t consume alcohol then 0.
12. Cough:
• If have cough disease then 1;
• If don’t have cough disease then 0.
13. Shortness of Breath:
• If have shortness of breath then 1;
• If don’t have shortness of breath then 0.
14. Hard to Swallow:
• If have difficulty swallowing then 1;
• If do not have Difficult Swallowing then 0
15. Chest Pain:
• If have chest pain then 1;
• If don’t have chest pain then 0.
16. Lung Cancer:
• If YES then 1;
• If NO then 0.
2.2 Research Stages and Hebbian
Algorithm Architecture
To achieve the research objectives, the following steps
were taken to complete the research. The following is
a picture of the stages of the research.
Figure 1: Research Stages.
Figure 2: Hebbian Architecture.
3 RESULTS AND DISCUSSION
The simulation results of Hebbian’s algorithm testing
to predict lung cancer in smokers with 4 forms of in-
put and output patterns produce accuracy that is not
much different from one another. The following is
the result of the summation of the test with 4 forms of
input and output patterns.
3.1 Bipolar Input and Output Patterns
The results of the simulation test using the Hebbian
algorithm to predict lung cancer with bipolar input
and output patterns yield an accuracy of 60%. The
following is the simulation result of lung cancer pre-
diction testing in smokers.
ICAISD 2023 - International Conference on Advanced Information Scientific Development
278

Table 2: Test Simulation Results with Bipolar Input and
Output Patterns.
x1 x2 x3 ... x15 y=f(net) t Result
1 1 -1 ... 1 1 1 True
1 1 1 ... 1 1 1 True
-1 1 -1 ... 1 1 -1 False
1 1 1 ... 1 1 -1 False
-1 1 -1 ... -1 1 -1 False
-1 1 -1 ... -1 1 1 True
1 -1 1 ... 1 1 1 True
-1 -1 1 ... -1 1 1 True
-1 1 1 ... -1 1 -1 False
1 -1 1 ... 1 -1 1 False
... ... ... ... ... ... ... ...
-1 1 1 ... -1 1 1 True
-1 1 -1 ... -1 1 -1 False
3.2 Binary Input and Output Patterns
Simulation test results using the Hebbian algorithm to
predict lung cancer with binary input and output pat-
terns produce an accuracy of 65%. The following is
the simulation result of lung cancer prediction testing
in smokers.
Table 3: Test Simulation Results with Binary Input and Out-
put Patterns.
x1 x2 x3 ... x15 y=f(net) t Result
1 1 0 ... 1 1 1 True
1 1 1 ... 1 1 1 True
0 1 0 ... 1 1 0 False
1 1 1 ... 1 1 0 False
0 1 0 ... 0 1 0 False
0 1 0 ... 0 1 1 True
1 0 1 ... 1 1 1 True
0 0 1 ... 0 1 1 True
0 1 1 ... 0 1 0 False
1 0 1 ... 1 1 1 True
... ... ... ... ... ... ... ...
0 1 1 ... 0 1 1 True
0 1 0 ... 0 1 0 False
3.3 Patterns of Bipolar Input and
Binary Output
The results of simulation tests using the Hebbian al-
gorithm to predict lung cancer with bipolar input and
binary output patterns yield an accuracy of 60%. The
following is the simulation result of lung cancer pre-
diction testing in smokers.
Table 4: Test Simulation Results with Bipolar Input and Bi-
nary Output Patterns.
x1 x2 x3 ... x15 y=f(net) t Result
1 1 -1 ... 1 1 1 True
1 1 1 ... 1 1 1 True
-1 1 -1 ... 1 1 0 False
1 1 1 ... 1 1 0 False
-1 1 -1 ... -1 0 0 True
-1 1 -1 ... -1 0 1 False
1 -1 1 ... 1 1 1 True
-1 -1 1 ... -1 0 1 False
-1 1 1 ... -1 0 0 True
1 -1 1 ... 1 1 1 True
-1 1 1 ... -1 0 1 False
... ... ... ... ... ... ... ...
-1 1 1 ... -1 0 1 False
-1 1 -1 ... -1 0 0 True
3.4 Patterns of Binary Input and
Bipolar Output
The results of simulation testing using the Hebbian
algorithm to predict lung cancer with a bipolar input
pattern and binary output yield an accuracy of 65%.
The following is the simulation result of lung cancer
prediction testing in smokers.
Table 5: Test Simulation Results with Bipolar Input and Bi-
nary Output Patterns.
x1 x2 x3 ... x15 y=f(net) t Result
1 1 0 ... 1 1 1 True
1 1 1 ... 1 1 1 True
0 1 0 ... 1 1 -1 False
1 1 1 ... 1 1 -1 False
0 1 0 ... 0 1 -1 False
0 1 0 ... 0 1 1 True
1 0 1 ... 1 1 1 True
0 0 1 ... 0 1 1 True
0 1 1 ... 0 1 -1 False
1 0 1 ... 1 1 1 True
... ... ... ... ... ... ... ...
0 1 1 ... 0 1 1 True
0 1 0 ... 0 1 -1 False
Based on the simulation results of the Hebbian
algorithm testing to predict lung cancer for smok-
ers with binary input and output patterns producing
65% accuracy, with bipolar input and output patterns
producing 60% accuracy, with binary input and bipo-
lar output patterns producing 65% accuracy and with
bipolar input pattern and binary output give 60% ac-
curacy.
Comparison of the Hebbian Algorithm Based on Input and Output Patterns in the Prediction of Lung Cancer in Smokers
279

3.4.1 Comparison of Simulation Results of
Hebbian Algorithm Testing
After conducting a simulation test using the Hebbian
algorithm, the best accuracy value is Hebbian learn-
ing with binary input patterns, both with binary output
and with bipolar output, which produces an accuracy
value of 65
4 CONCLUSIONS
The conclusion of this study is that the Hebbian al-
gorithm can predict lung cancer in smokers with an
accuracy of 65% with binary input patterns and bi-
nary and bipolar output patterns. To produce better
test results for accuracy, it is necessary to study with
other algorithms.
REFERENCES
Bisri, H., Bustomi, M., and Purwanti, E. (2013). Image
classification of the lungs by histogram feature extrac-
tion and backpropagation neural networks. J. Sains
Dan Seni Pomits, 2(2):68–71,.
Gillett, M., Pereira, U., and Brunel, N. (2020). Characteris-
tics of sequential activity in networks with temporally
asymmetric hebbian learning. Proc. Natl. Acad. Sci.
U. S. A, 117(47):29948–29958,.
Golkar, S., Lipshutz, D., Bahroun, Y., Sengupta, A.,
and Chklovskii, D. (2020-12). A simple norma-
tive network approximates local non-hebbian learn-
ing in the cortex. Adv. Neural Inf. Process. Syst,
(NeurIPS):1–13,.
Illing, B., Ventura, J., Bellec, G., and Gerstner,
W. (2021). Local plasticity rules can learn
deep representations using self-supervised contrastive
predictions. Adv. Neural Inf. Process. Syst,
36(NeurIPS):30365–30379,.
Isomura, T. and Toyoizumi, T. (2019). Multi-context blind
source separation by error-gated hebbian rule. Sci.
Rep, 9(1):1–13,.
Journ
´
e, A., Rodriguez, H., Guo, Q., and Moraitis, T. (2022).
Hebbian deep learning without feedback. Available:.
Kwessi, E. (2022). Strong allee effect synaptic plasticity
rule in an unsupervised learning environment. Avail-
able:.
Liu, F., Sekh, A., Quek, C., Ng, G., and Prasad, D.
(2021). Rs-herr: a rough set-based hebbian rule re-
duction neuro-fuzzy system. Neural Comput. Appl,
33(4):1123–1137,.
Magotra, A. and Kim, J. (2020). Improvement of hetero-
geneous transfer learning efficiency by using hebbian
learning principle. Appl. Sci, 10(16).
Muscoloni, A. and Cannistraci, C. (2021). Short note
on comparing stacking modelling versus cannistraci-
hebb adaptive network automata for link prediction in
complex networks. Preprints, (May).
Najarro, E. and Risi, S. (2020-12). Meta-learning through
hebbian plasticity in random networks. Adv. Neural
Inf. Process. Syst.
Napitupulu, S. and Situmorang, Z. (2020). Optimization of
giving employee craft assessment using artificial neu-
ral network with hebb algorithm,”iop conf. Ser. Mater.
Sci. Eng, 725(1).
Napole, C., Barambones, O., Calvo, I., and Velasco, J.
(2020). Feedforward compensation analysis of piezo-
electric actuators using artificial neural networks with
conventional pid controller and single-neuron pid
based on hebb learning rules. Energies, 13(15):1–16,.
Nicola, W. and Clopath, C. (2019). A diversity of in-
terneurons and hebbian plasticity facilitate rapid com-
pressible learning in the hippocampus. Nat. Neurosci,
22(7):1168–1181,.
Osakabe, Y., Sato, S., Akima, H., Kinjo, M., and Sakuraba,
M. (2021). Learning rule for a quantum neural net-
work inspired by hebbian learning. IEICE Trans. Inf.
Syst, D(2):237–245,.
Pedersen, J. and Risi, S. (2021). Evolving and merging heb-
bian learning rules: Increasing generalization by de-
creasing the number of rules, volume 1. Association
for Computing Machinery.
Pogodin, R. and Latham, P. (2020-12). Kernelized informa-
tion bottleneck leads to biologically plausible 3-factor
hebbian learning in deep networks. Adv. Neural Inf.
Process. Syst.
Prasetio, R. and Susanti, S. (2019). Prediction of life ex-
pectancy in lung cancer patients after thoracic surgery
using boosted k-nearest neighbor. J. Responsif,
1(1):64–69,. Available:.
Qin, Y. and Duan, H. (2020). Single-neuron adaptive hys-
teresis compensation of piezoelectric actuator based
on hebb learning rules,”micromachines.
Setyadi, Y., Asror, I., and Wibowo, Y. (2020). Prediction of
life expectancy after thoracic surgery in patients with
lung cancer using genetic algorithm methods for fea-
ture selection and na
¨
ıve bayes classifier. e-Proceeding
Eng, 7(20):8349–8360,.
Solikhun, M., Safii, M., and Zarlis, M. (2020a). Back-
propagation network optimization using one step se-
cant (oss) algorithm. IOP Conf. Ser. Mater. Sci. Eng,
769(1).
Solikhun, M., Safii, M., and Zarlis, M. (2020b). Resilient
algorithm in predicting fertilizer imports by major
countries. IOP Conf. Ser. Mater. Sci. Eng, 769(1).
Tsai, M. (2021). Photoactive electro-controlled visual
perception memory for emulating synaptic metaplas-
ticity and hebbian learning. Adv. Funct. Mater,
31(40):13–14,.
ICAISD 2023 - International Conference on Advanced Information Scientific Development
280