Music Genre Classification Using Ga

Music Genre Classification Using Ga-Induced Minimal Essay In addition to an enhanced performance, our model can also reduce the computational load for ill-formed sets and has the flexibility to incorporate trade-offs between efficiency and computational load. We finally demonstrate that the modified GA is capable of bringing about an 80% reduction in the feature space dimension at similar classification rates. Index Terms?Genetic Algorithms, Feature-set Reduction, SAVE, ann. Classifier, Genre Classification Music genre is frequently used as the query of choice while browsing through the music databases over the Internet. While research into acquiring suitable features and classifiers for genre classification is abundant(, , , ), works involving selection of relevant feature-sets particular to the task at hand are sparse at best. Some examples of such work would be , , where techniques of PICA, OFFS and BBS have been used effectively. In our work, we investigate the possibility of a relevant and minimal feature-set-selection with the help of Genetic Algorithms (GA). GAS have been used before by and to efficiently classify music genres. We will write a custom essay on Music Genre Classification Using Ga-Induced Minimal specifically for you for only $16.38 $13.9/page Order now While has shown pretty impressive results (a 90% lactations rate) using GA, 109 musical features and a hierarchical classifier, has been able to get almost 60% classification rate with GA, a set of classifiers like ann., SAVE and a feature set of 30 features. These works, though impressive, aim only for an optimal feature-set devoid of noisy features(which tend to misclassifying samples). Furthermore, they use a primitive GA fitness function, biz. The hit-rate, for the best feature-set selection, which might converge to a local minima instead of a global one . In this work, we not only find the optimal features from a set of given features for given classification problem, but we also minimize the number of features required for the task. Using a modified fitness function, that also eliminates the local-minima problem of the standard hit-rate one, we bring about a more than 80% reduction in the original feature set, and a 50% reduction in the feature-set derived from the original feature-set by the hit-rate-only fitness function. To reiterate, we havent proposed a new method for classification, or a new method for feature generation. We assume that the feature set and the classifier are defined by Music Genre Classification Using Ga-launched Minimal Feature-Set By rampant feature reduction techniques have already been employed to reduce the original feature-set, can we further reduce the number of features with minimal loss in performance? thighs the question we are going to handle in this paper. The paper is organized as follows. In Section II, we give a brief description of the established GA and feature selection methods in literature, and propose our modification. Then in Section Ill, we describe the dataset, the feature-set and classifiers we are using for the experiment. Then in Section V, we discuss the findings of our test and validate our claims. Then we conclude our discussion in Section V. II. P REPOSED M DEL Given a set of possible features and a classifier for a classification task, our objective is to derive an optimal features devoid of noisy features with an aim of minimizing its cardinality. Using GAS to derive near-optimal feature-subsets was brought to light by , and was further investigated in , . In the present work, extending and , we propose a modified GA to derive a near-optimal-UCM-minimal featureless for music genre classification. GAS emulate human evolutionary characteristics, and follow the survival-of-the-fittest paradigm. In an personalized model of GA, we have a population of chromosomes (essentially a bit string), which are assigned fitness values (their relative worth) based on the task at hand. Fit individuals/chromosomes mate with each other, and that is simulated by crossover, in which they exchange parts of their chromosome. .u17b01f798ef07d0bb8b4b702e0d3702e , .u17b01f798ef07d0bb8b4b702e0d3702e .postImageUrl , .u17b01f798ef07d0bb8b4b702e0d3702e .centered-text-area { min-height: 80px; position: relative; } .u17b01f798ef07d0bb8b4b702e0d3702e , .u17b01f798ef07d0bb8b4b702e0d3702e:hover , .u17b01f798ef07d0bb8b4b702e0d3702e:visited , .u17b01f798ef07d0bb8b4b702e0d3702e:active { border:0!important; } .u17b01f798ef07d0bb8b4b702e0d3702e .clearfix:after { content: ""; display: table; clear: both; } .u17b01f798ef07d0bb8b4b702e0d3702e { display: block; transition: background-color 250ms; webkit-transition: background-color 250ms; width: 100%; opacity: 1; transition: opacity 250ms; webkit-transition: opacity 250ms; background-color: #95A5A6; } .u17b01f798ef07d0bb8b4b702e0d3702e:active , .u17b01f798ef07d0bb8b4b702e0d3702e:hover { opacity: 1; transition: opacity 250ms; webkit-transition: opacity 250ms; background-color: #2C3E50; } .u17b01f798ef07d0bb8b4b702e0d3702e .centered-text-area { width: 100%; position: relative ; } .u17b01f798ef07d0bb8b4b702e0d3702e .ctaText { border-bottom: 0 solid #fff; color: #2980B9; font-size: 16px; font-weight: bold; margin: 0; padding: 0; text-decoration: underline; } .u17b01f798ef07d0bb8b4b702e0d3702e .postTitle { color: #FFFFFF; font-size: 16px; font-weight: 600; margin: 0; padding: 0; width: 100%; } .u17b01f798ef07d0bb8b4b702e0d3702e .ctaButton { background-color: #7F8C8D!important; color: #2980B9; border: none; border-radius: 3px; box-shadow: none; font-size: 14px; font-weight: bold; line-height: 26px; moz-border-radius: 3px; text-align: center; text-decoration: none; text-shadow: none; width: 80px; min-height: 80px; background: url(https://artscolumbia.org/wp-content/plugins/intelly-related-posts/assets/images/simple-arrow.png)no-repeat; position: absolute; right: 0; top: 0; } .u17b01f798ef07d0bb8b4b702e0d3702e:hover .ctaButton { background-color: #34495E!important; } .u17b01f798ef07d0bb8b4b702e0d3702e .centered-text { display: table; height: 80px; padding-left : 18px; top: 0; } .u17b01f798ef07d0bb8b4b702e0d3702e .u17b01f798ef07d0bb8b4b702e0d3702e-content { display: table-cell; margin: 0; padding: 0; padding-right: 108px; position: relative; vertical-align: middle; width: 100%; } .u17b01f798ef07d0bb8b4b702e0d3702e:after { content: ""; display: block; clear: both; } READ: Services Marketing: CRM and the Music Industry EssayTo account for sudden characteristic changes in some individuals of a population, mutation is simulated through bit flipping random parts of the chromosome. The fit individuals are selected room a population using some selection rule and the next generation consists of these fit individuals whereas unfit ones are discarded. With this approach, the GA claims that at the end of hundreds of generations, only the best chromosomes for the task at hand will remain, and that is the assumption based on which best features are selected. The crossover rate, mutation rate, selection rule, fitness function etc. Are determined by the problem at hand. For an n-dimensional feature space, we choose n-bit chromosomes. Following , a bit-value of 1 means that the feature-dimension is considered for classification, and value of O means that the feature is eliminated from consideration. 1 The traditional fitness-function is the hit-rate (HER), I. E. The fraction of train-set samples that were correctly classified using the selected features. But such a fitness function has a tendency to converge to a local minima, and it also disregards the objective of creating the maximum separation between the classes under consideration. So, following , we use the following fitness function: F tiniest = HER + y(CD/p) where CD is the class distance, p is the total number of samples, and y is a scaling constant. The class distance is defined as follows, p p Ill. E EXPERIMENT A. Feature Set We used audio features described in . Each music file is represented in a 74-dimensional feature space.