This study investigates the application of a modified Negative Selection Algorithm (NSA), derived from

principles of the human immune system, to enhance music genre classification. NSA’s threshold-based similarity

adaptability and robustness in handling genre variability, especially in cross-domain music datasets. Beyond

technical contributions, this study emphasizes the importance of understanding musical features that define genre

identity. By offering a biologically inspired, threshold-sensitive model, the research contributes to the

development of intelligent, interpretable systems for multimedia classification. The approach supports more

accurate music categorization, which has implications for recommendation systems, digital archiving, and cross-

cultural music analysis. This work bridges computational intelligence and music analysis, offering a novel

perspective on immune-inspired learning for content classification. It reinforces the potential of NSA as a

practical and scalable tool for genre recognition in diverse musical contexts.

human perceptual identification of rhythm, timbre, and melody. In the content-based music genre categorization,

feature representation and extraction are crucial, as they offer the foundation for distinguishing between musical

genres and structures [1].

Early efforts on genre classification generally focused on low-level acoustic characteristics and traditional

machine learning classifiers, such as k-nearest neighbour (KNN), Naïve Bayes, and Support Vector Machines

[2]. More recently, deep learning models using convolutional and recurrent neural architectures have improved

accuracy by learning complex, hierarchical representations directly from audio input [2][3]. Despite these

developments, the need for interpretable, adaptive, and physiologically inspired models remains crucial.

The Artificial Immune System (AIS), modelled on human immunological processes such as recognition and

adaptability, offers an alternative technique for pattern classification through its Negative Selection Algorithm

in music information retrieval (MIR) reflect a transition toward multimodal and cross-cultural viewpoints.

Studies such as [5] provide cross-modal retrieval frameworks merging text-based semantic descriptions and

music embeddings to facilitate similarity learning. Moreover, cross-cultural experiments have demonstrated

significant perceptual differences in genre similarity judgments between Western and non-Western listeners [6],

suggesting that classifier performance can vary based on cultural bias and dataset composition.

These latest findings highlight the usefulness of adaptive and interpretable models like AIS, which might

possibly accommodate such variety in musical structure and perception.

Content-based music genre classification has progressed from feature-based heuristics to data-driven and

biologically inspired algorithms. Old methods used timbre, pitch, and rhythm as the main ways to describe music

approaches [10] examine song portions and transitions, yielding an elevated comprehension of musical form that

transcends basic signal patterns. These methods are like the conceptual design of AIS, where feature

representation and hierarchical detection are like the immune system's ability to recognize complicated stimuli.

Research that looks at different cultures and focuses on melody also shows that there is a rising demand for

evaluation standards that include everyone. [11] presented a melody-aware similarity dataset for cross-domain

plagiarism and similarity detection, highlighting the variations in melodic contours across cultural settings.

Research on cultural variety [12] illustrates the impact of dataset bias on model generalization, corroborating

findings in this study that categorization accuracy differs markedly between Western and Asian music

collections.

Deep models are the most common kind of models in literature right now [13], but hybrid techniques, like the

applications based on the mechanisms of recognising self or non-self-cells.

The negative selection algorithm follows the distinction of self or non-self-cell processes in the thymus, which

is achieved by T-cells having receptors on their surfaces, operating as a detector to identify the antigens or the

foreign proteins. These receptors are created by a process called the pseudo-random genomic re-arrangement

during the production of T-cells in the thymus [15]. They subsequently go through a censoring process or the

negative selected cells task. Inside the thymus, the T-cells that recognise and react to the self-cells are eliminated

with the consequence that only those cells which do not bind themselves to the self-proteins are allowed to leave

the thymus. These developed cells circulate in the bloodstream throughout the whole-body hunting for foreign

The method normally works by deleting self-cells and randomly turns the detected non-self-cells into detectors.

These detectors then will distinguish a new batch of non-self-cells and kill the detected self-cells. Two stages

are involved in the process: filtering and monitoring. Censoring is the procedure of creating detectors at random,

while monitoring is the process of comparing the detectors with cells to identify non-self-cells. If matched, the

system. It is the major component of the thymus that identifies the cells which are alien to the human body. From

a pattern recognition perspective, the detectors play key roles in detecting the patterns. That is why, researchers

are making the effort to analyse the process of manufacturing the detectors and comparing them in terms of space

complexity and time to manufacture competent detectors.

As the objective of this research is to accurately categorize music genres, several adjustments were made to the

NSA so that it can be utilized to tackle the low performances of earlier music genre classification difficulties.

The suggested updated NSA is dubbed the updated AIS-based music genre classifier.

The biggest difficulty with the NSA is the technique of producing random detectors which can impair the

accuracy of categorization. The next section will discuss the adjustment that was made in generating the detectors

mechanism.

As the current research is investigating comparable patterns in the music genre, the recognition will need precise

and dedicated detectors to accomplish the operation [15]. If the assumptions and findings of the previous

The created detectors ought to be efficient enough so that the detecting process will yield good performance in

any identifying task. It is not only focusing on the detectors, but the process to generate the detectors need to be

extremely effective to develop effective detectors as well. The generalization idea would not be an emphasized

and monitoring modules play essential roles as they were the fundamental procedures of the Modified AIS-based

classifier.

The detectors are generated purely based on how many patterns they should recognize. This is to ensure that

there will be a complete set of antibodies that are needed to recognize the antigens. The algorithm contains a

procedure that highlights the XOR-operation in creating the detectors and this process is termed as a detector

devoted generated process. This procedure not only eliminates the random process, and the generation process

is done in a timely manner, but it can also ensure the generated detectors are appropriate in number for pattern

recognition. During the classification process, the detectors are generated from the training data (as this research

utilizes the supervised learning method) which later will be utilized to test the testing data to find the

classification accuracy.

Before any songs can be classified according to genres, the songs feature vectors first need to be translated into

binary strings. At this point, transformation is highly significant as the similarity matching techniques that are

employed use the binary strings to calculate the classification accuracy based on the song genres. Since the

features are then classified and applied to two important aspects of the AIS.

In the censoring process, the song feature vectors are transformed into binary strings where they go through a

detector dedicated generation process and are then stored as song detectors. These detectors act as antibodies as

During the monitoring process, the similarity matching procedure is applied by comparing the song detectors

with the song antigens to find out which songs are similar, and those similar songs are identified as matched.

procedure, four separate binary matching approaches are utilized to obtain the matched song similarity

percentage. The similarity percentage is used to determine whether detectors and antigens are matched by using

a threshold value.

The methodology incorporates the Artificial Immune System (AIS) paradigm into a supervised learning

framework for music genre classification. Inspired by the immune system’s ability to recognize “self” and “non-

self” things, the algorithm adapts the Negative Selection Algorithm (NSA) by strengthening its detector

generation and affinity evaluation processes. The two primary modules: Censoring and Monitoring are revised

to ensure robust categorization and minimize randomization in detector creation.

implementations could combine precision, recall, and F1-score metrics where, lately stressed in deep MIR

research [14] once confusion matrices become available.

the selected features are translated and represented by binary strings (for example, feature vector -3.4523123 is