Analysis of Isotonic Calibration on Gaussian Naïve Bayes Performance for Guitar Chords Classification

Music recognition and recommendation is an area of machine learning which is gaining popularity by the day amongst interdisciplinary researchers. Guitar chord classification problem has been making rounds in machine learning enthusiasts for music recognition and recommendation models. Fundamentals of guitar playing and learning often includes the knowledge about guitar strings and chords etc. Going one step deeper, the domains like music recognition and music recommendation systems are evolving at a very fast pace. In this paper, we have worked on Guitar chord classification problem which has been explored and experimented in more than one dimension for music recognition and recommendation models. The preprocessing of guitar chords audio into numerical data is computationally extensive. During the preprocessing stage the harmonics are generated in large numbers and the effect of higher order harmonics is explored in some previous works. In this paper we have evaluated the Brier score losses of the Gaussian Naïve bayes algorithm for guitar chord classification problem for two sets of harmonics count. Further, the isotonic calibration is done on the classification problem and the experimental results are obtained and presented in the results and discussion section. Isotonic calibration of Gaussian Naïve bayes shows great promise in the performance where it gives improved values of Brier score for both harmonics inclusion cases. The results are improved by up 5% even when the harmonics are included upto 9<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> position.