2024 International Conference on Intelligent Algorithms for Computational Intelligence Systems (IACIS)

A Chaotic Gradient-based Optimization with

Support Vector Machine for Chinese Folk Music
Classification
2024 International Conference on Intelligent Algorithms for Computational Intelligence Systems (IACIS) | 979-8-3503-6066-0/24/$31.00 ©2024 IEEE | DOI: 10.1109/IACIS61494.2024.10721674

Fan Zhang
Xi’an Shiyou University
Xi’an, China
[email protected]

Abstract—Nowadays, Chinese folk music classification has a music. Using Machine Learning (ML) technologies [4-5],
huge attraction on the researches end where traditional Chinese researchers aimed to preserve the cultural heritage of Chinese
folk music classification has difficulty with feature selection, and folk music by developing powerful and effective
current Machine Learning (ML) algorithms have limitations categorization systems. ML is a subfield of AI which focuses
such as poor classification accuracy. Conventional music on creating models that learn from data and utilize that
classification algorithms struggle to accurately classify Chinese knowledge for classification. SVM is a popular ML technique
folk music due to its sensitive rhythms and tone changes. The for regression and classification issues. However, without
goal is to develop algorithms that can accurately recognize and adequate feature selection, SVMs have trouble in classifying
classify diverse folk tunes in the Chinese music collection. So,
both high-dimensional and nonlinear data. So, the CGBO
Chaotic Gradient-Based Optimizer (CGBO) is introduced that
merges chaotic maps with GBO searching rounds to find the
combines chaotic maps with GBO search rounds to find the
most discriminating features and finally for folk music most discriminating features. This study highlights the
categorization, the Support Vector Machine (SVM) algorithm is effectiveness are utilized to protect cultural heritage, and it
used. In this research to evaluate the performance of the represents a significant development in Chinese folk music
proposed CGBO-SVM, input data is created by self as set of categorization systems. Kostrzewa et al [6] suggested that
ethnic music data, this data collection includes 10 different music genre recognition be done by using the capabilities of
genres. Every genre includes minimum count of 100 samples, large ensembles of neural network classifiers. Researchers
and every sample lasting 30 secs with sampling rate of 22050 Hz, have made significant contributions to the progress of
while the One Note detector technique is utilized to effectively automated musical genre identification by creating
extract features. The experimental results reveal that the uncommonly utilised forms of ensembles. By incorporating
suggested CGBO-SVM achieve 99.96% of classification large ensembles, researchers get another approach for
accuracy and performs better than existing approaches such as enhancing the classification outcome without requiring
K-Nearest Neighbor (KNN) and Bi-Directional Long Short significant design or programming work. A limitation might
Term Memory (Bi-LSTM). be higher calculation time as well as demand for computer
resources.
Keywords—chinese folk music classification, machine
learning, chaotic gradient-based optimizer, support vector Wijaya et al. [7] proposed a customized BiLSTM model
machine, one note detector. with Mel Frequency Cepstral Coefficients (MFCC) feature
extraction. BiLSTM was a neural network design that enables
I. INTRODUCTION the model to acquire patterns in both directions. To get
As similar to art, Music communicates people's ideas, normalized input, preparatory techniques such as silent
emotions, and lifestyles, as well as promote emotional and portion removal and stretching are undertaken. The model was
spiritual well-being [1]. The emergence of Music has quite excellent at categorizing positive occurrences of the
suddenly rise and become popular in people's circumstances. "pop" class; but, the model tends to overlook certain positive
As science and technology advanced, an increasing number of examples that should be categorized; this might be owing to
individuals preferred to listen music over the Internet. As a the relatively short sample size, which was just six records.
result, determining which music people want to listen to from Zhang [8] proposed a unique technique that uses music
a large collection of music has become more crucial, and extraction as well as KNN to solve the issue of poor accuracy
music identification and categorization have received in existing methods. As a consequence, a music categorization
increased attention. Particularly, the examples of Chinese folk module based on a KNN's one-dimensional convolution was
songs are Instrumental music, songs, and opera [2] which suggested , which paired with a two-way KNN network. The
comprises a diverse set of essential forms and approaches output was weighted differently to properly reflect the music
based on Chinese culture. Despite its broad reach and style features. The tonal feature was employed to achieve
historical importance, academics encounter a number of performance second only to the suggested approach, however
challenges when classifying Chinese folk music [3]. The the melody element has a higher influence on performance.
intricate structure of Chinese folk music is a difficulty for Ceylan [9] presented a Convolutional Neural Network (CNN)
traditional algorithms, thus resulting in erroneous to categorize music genres, based on the prior successful
classification outputs. Folk music relies heavily on musical findings. Instead of utilising visual characteristics or
instruments where it may be played individually or in an representations, music segments in the dataset were
ensemble, and various combinations of musical instruments represented using MFCC. We obtain MFCCs by
create diverse kinds of instrumental music. The purpose of this preprocessing the dataset's music pieces, then train a CNN
study is to bridge the gap between old forms of Chinese folk model using the MFCCs and assess the model's success using
music and modern technology by classifying Chinese folk the testing data. A definite decision cannot be formed since
certain instances in deep learning remain unexplained. Elbir

979-8-3503-6066-0/24/$31.00 ©2024 IEEE

Authorized licensed use limited to: North South University. Downloaded on October 26,2024 at 09:35:45 UTC from IEEE Xplore. Restrictions apply.
[10] proposed DeepMUSIC, a DL system for multiple signal • The self-created ethnic music data collection consists
classification. Each CNN acquires the MUSIC spectra of its of ten genres, each with 100 audio samples lasting 30
own angular subregion. The algorithm created a non-linear seconds where the sampling rate is 22050 Hz, and from
link between sensor data and the angular spectrum. While the data features are extracted using the Note onset
DeepMUSIC includes numerous networks in comparison to detector technique.
Multi-Layer Perceptron (MLP), it gives less computing time
because DeepMUSIC has several convolutional layers rather • Finally, for effective Chinese folk music
than fully linked layers, which require greater complexity. categorization, the SVM is used which classifies
optimum features.
The primary contributions of the research are as follows:
The remaining of the paper is: The proposed methodology
• The CGBO combines chaotic maps with GBO is briefly explained in section 2 whereas the experimental
iterations to select discriminating characteristics and results are detailed in section 3. At last, the conclusion of the
improve accuracy in Chinese folk music classification. overall research is presented in section 4.
II. PROPOSED METHODOLOGY

Fig. 1. Block diagram of the proposed methodology

In this research, CGBO that combines chaotic maps with Through integrating greatly associated characteristics/
searching iterations of the GBO for selecting most features with additional characteristics, ML techniques
discriminating features to achieve high performance in improve classification accuracy. Eventhough if two
Chinese folk music classification. As shown in Figure 1, the characteristics have a great connection, only one is employed
overall procedure followed in this research is described in the to correctly describe the data.
following sections clearly.
1) Gradient-Based Optimizer (GBO): GBO [12] is a
A. Data Collection strategy for addressing hard optimisation issues that
In this research to evaluate the performance of the combines population-based and gradient approaches. The
proposed CGBO-SVM, input data is created by self as ethnic GBO method uses Newton's technique to control the
music data this data collection includes 10 different genres. direction of the search agent while it explores the problem
Every group includes minimum count of 100 samples, where space as shown in Eq (1). The Gradient Search Rule (GSR)
an every sample lasting 30 secs. Whereas 100 audio samples along the Locale Escaping Operator (LEO) are two key
sampling rate is 22050 Hz [11]. The collected input samples factors of GBO method.
are given to one note extraction technique for feature
0,1
extraction.
(1)
B. Feature Extraction
Note onset detector extraction is the process of finding the 2) GSR Phase : GBO technique begins with group of
exact times in audio data that musical notes originate. initial solutions it updates each agent's location in a gradient-
This method monitors variations in spectral content over time. based location. The crucial factor is employed to establish a
Notation onsets are associated with fast signal shifts balance between investigating substantial search space
(Melspectrum features) that are captured by note onset regions as well as exploitation to get near-optimal solutions.
extraction. Note onset times are accurately determined by
using time-domain properties including short-time energy and
amplitude envelope where the extracted features are given as (2)
worst best
inputs to CGBO for feature selection.
GBO does a randomized exploration, which involves
C. Feature Selection (FS) locating local optima utilizing random behaviour. The above
From the extracted features, FS allows picking a feature equation, Eq. (2), demonstrates how to calculate the random
subclass having necessary information and redundancy free offset that represents the variance among optimal answer and
thus enhances categorization performance substantially.

Authorized licensed use limited to: North South University. Downloaded on October 26,2024 at 09:35:45 UTC from IEEE Xplore. Restrictions apply.
random chosen answer. In exploration stage, additional are employed which helps to prevent local optima as well as
random number is in eqs (3-5): early convergence, as well as; SVM is subsequently utilized

2 ! !
for classification. The suggested technique's solutions must be
(3) examined at every iteration to verify that they function as
predicted. The fitness function (6@AB) of CGBO is described
! "# $% & $% )# *+"
'( '( in Eq (7):
(4)
C%D E$$ 67 FD%@ 6@AB ! #
|H|
' |I|
# # # # &1 ) * +
(7)

,
(5)
SVM is a quadratic method employed in this research to
accurately classify Chinese folk music. The SVM classifier's
3) LEO Phase: LEO is introduced to improve the purpose is to generate "good" separation hyperplanes between
suggested GBO’s efficacy in solving complex issues where classes in a high-dimensional feature space in a
the LEO can successfully update the solution's location. It computationally efficient way. The input vectors are turned
helps GBO break out of local optimum spots as well as into a high-dimensional feature space by using nonlinear
accelerate convergence. As shown in Eq (6), LEO gives a transformation as well as linear separation. To produce a
solution with greater efficacy by merging many solutions, nonlinear SVM classifier, the inner product is replaced by a
kernel function, as indicated in Eq (8),
If rand < .
6 8 $J :∑! LM N 8 , O A;
6 7 8best 7 89
⎧
(8)

⎪ 6 :7' 2 1 ;
⎪
7 8< 8< /2, if rand < 0.5
A SVM classifier's fundamental form is as in Eq (9):
/01
⎨ 6 7 8best 7 89 P O QRS O A
(6)

⎪ 6 :7' 2 1 ;
(9)
⎪
⎩ 7 8< 8< /2, otherwise space Φ O .
The feature mapping function result is the feature

From (7) and (8), the test data O of SVM is in eq (10),

End
Specifically, the CGBO’s stages are carried out as: To
preserve a balance among exploitation & exploration, CMs
P O ∑V
WXY !WX LWX S ZWX S O
R
A ∑V
WXY ∑WX LWX N ZWX , O A (10)

From this, it clearly demonstrates that the SVM obtain The performance of the CGBO is analyzed with the
higher classification accuracy in Chinese folk music traditional approaches such as Bat Optimization (BOA),
classification. Whale Optimization Algorithm (WOA) to determine its
efficiency whereas the obtain results are tabulated in table 1.
III. EXPERIMENTAL RESULTS The analysis is conducted using the considered performance
In this section, using the MATLAB 2020 software, Intel measures such as recognition accuracy and Error rate.
i7, 8GB RAM, 32 bit processor, the performance of the From the above table, it clearly demonstrates that the
proposed CGBO-SVM was tested. The purpose of the proposed method obtains a higher recognition and data loss of
simulation experiments is to confirm that the CGBO-SVM 99.96% and 1.17%. It clearly demonstrates that the CGBO
approach works effectively. To evaluate the proposed CGBO- outperforms the traditional approaches. This superior
SVM such as classification accuracy and error rate where the performance is achieved through the utilization of CGBO for
mathematical equations is given in Equations (11 and 12). feature selection process. Thus, the CGBO-SVM model

[\ $$%6%F D%@ ]FF7 FL

R^ RV
shows high level of effectiveness in Chinese folk music
R^ RV _^ _V
(11) classification.

` @ DE
_^ _V
The comparison analysis between proposed CGBO with
^ V
(12) conventional models is conducted. The enhanced performance

Where, ab-True Positive, Cb – False Positive, ac-True

of the CGBO-SVM is determined by comparing it with the

Negative, Cc- False Negative.

conventional approaches such as KNN [8], Bi-LSTM [7] to
determine its efficiency and the obtained results are tabulated
in table 2. The comparison is conducted using classification
TABLE I. PERFORMANCE ANALYSIS OF THE PROPOSED CGBO WITH accuracy.
TRADITIONAL APPROACHES
TABLE II. COMPARISON EVALUATION OF THE PROPOSED WITH
Methods Classification Accuracy (%) Error rate (%) EXISTING APPROACHES
GBO 93.54 13.07
BOA 95.55 8.16 Models Classification Accuracy (%)
WOA 96.22 4.34 KNN [8] 95.4
CGBO 99.96 1.17 Bi-LSTM [7] 94.60
CGBO-SVM 99.96

Authorized licensed use limited to: North South University. Downloaded on October 26,2024 at 09:35:45 UTC from IEEE Xplore. Restrictions apply.
 From Table 2, it clearly demonstrates that the CGBO- [8] Satpathy, R.B. and GP, R., 2024. Conformal eight‐port dual band
SVM obtains a higher accuracy of 99.96% and outperforms antenna with switchable radiation pattern for 5G enabled on‐body
wireless communications. Microwave and Optical Technology Letters,
the existing works. Whereas the KNN [8] and Bi-LSTM [7] 66(1), p.e33910.
obtains an 95.4% and 94.60% respectively. This superior [9] Ceylan, Can, Hardalaç, Kara, and Firat. "Automatic music genre
performance is achieved through the utilization of CGBO for classification and its relation with music education." World Journal of
feature selection process. Thus, the CGBO-SVM model Education 11, no. 2 (2021): 36-45.
shows high level of effectiveness in Chinese folk music [10] Elbir, Ahmet M. "DeepMUSIC: Multiple signal classification via deep
classification learning." IEEE Sensors Letters 4, no. 4 (2020): 1-4.
[11] Ning, Qinliang, and Shi. "Artificial neural network for folk music style
A. Discussion classification." Mobile Information Systems 2022, no. 1 (2022):
In this research, a CGBO method is proposed for the 9203420.
SVM's hyperparameter optimization to achieve superior [12] Elminaam, Salama, Ibrahim,. Houssein, and Elsayed. "An efficient
chaotic gradient-based optimizer for feature selection." IEEE
classification accuracy. A set of self-created music data Access 10 (2022): 9271-9286.
collection is used to evaluate the CGBO-SVM and One Note
detector approach is used to efficiently extract features from
the input data whereas the advantages of the proposed
approach are discussed. The following results demonstrate the
efficiency of the CGBO-SVM in successfully forecasting the
data patterns. The CGBO-SVM obtains a higher accuracy of
99.96% indicating better overall performance contrasted to
traditional approaches whereas the KNN, Bi-LSTM obtains
95.4% and 94.60% respectively. This superior performance is
achieved through the utilization of CGBO for feature selection
process. Thus, the CGBO-SVM model shows higher
performance in Chinese folk music classification
IV. CONCLUSION
The main objective of this research is to create model for
precise classification of Chinese folk music. So, a CGBO is
proposed for feature selection to select optimal features from
the extracted feature, thus helps SVM to achieve superior
classification accuracy of Chinese folk music. A self-created
music collection is used to evaluate the CGBO-SVM and One
Note detector approach is used to efficiently extract features
from the input data. The CGBO-SVM obtains higher
classification accuracy of 99.96%. Thus, the CGBO-SVM
model shows high level of effectiveness in Chinese folk music
classification. The future work will be focused on utilizing
hybrid optimization techniques to improve the recognition
accuracy of the DL models.
REFERENCES
[1] Kobilova, Ezoz, Sobirova, and Najmiddinov. "The importance of
music education in the formation of musical culture and
spirituality." Academicia: An International Multidisciplinary Research
Journal 11, no. 1 (2021): 698-703.
[2] Lihan, Liu, and Thothum. "A Comparative Study of Piano
Accompaniment in Chinese Folk Songs and Chinese Modern Songs."
PhD diss., Mahasarakham University, 2020.
[3] Tang, Jing, and Phiphat Sornyai. "The cultural treasures of Baima
Tibetan folk songs in Gansu Province, China, as a resource for literacy
education in Chinese music History." International Journal of
Education and Literacy Studies 11, no. 3 (2023): 234-243.
[4] Liu, Hui, Jiang, Gamboa, Xue, and Schultz. "Bell shape embodying
zhongyong: The pitch histogram of traditional chinese anhemitonic
pentatonic folk songs." Applied Sciences 12, no. 16 (2022): 8343.
[5] Cai, Jiandong. "Informatization integration strategy of modern popular
music teaching and traditional music culture in colleges and
universities in the era of artificial intelligence." Applied Mathematics
and Nonlinear Sciences (2024).
[6] Kostrzewa, Daniel, Ciszynski, and Brzeski. "Evolvable hybrid
ensembles for musical genre classification." In Proceedings of the
Genetic and Evolutionary Computation Conference Companion, pp.
252-255. 2022.
[7] Wijaya, Niar, and Muslikh. "Music-genre classification using
Bidirectional long short-term memory and mel-frequency cepstral
coefficients." Journal of Computing Theories and Applications 1, no. 3
(2024): 243-256.

Authorized licensed use limited to: North South University. Downloaded on October 26,2024 at 09:35:45 UTC from IEEE Xplore. Restrictions apply.