Mar 2022 - May 2022
• Using 6 Machine Learning classifiers - Naive Bayes, Logistic Regression, K-Nearest Neighbours, Random kForest, Stochastic Gradient Descent and Cross Gradient Booster (XGBoost) to solve the multi-class classification problem.
• Ranking the performance of different classifiers on normalised and non-normalised data based on their precision, recall and F-measure.
• Using inter-feature, feature-targeted correlation and feature selection methods like Recursive Feature Extraction to minimise the spectral features of different musical genres.
• Tuning each model and fitting it to the train data using k-fold cross-validation and RandomizedSearchCV.
Problem Statement: The objective is to identify the genre of the song given a set of spectral features
Data: GTZAN Dataset
Tasks: Identify potential features of individual genres to identify the types of music genres. Design novel feature selection framework or use the existing techniques with proper justifications to identify salient features of the given data. Report the performance of different classification techniques on the training data to demonstrate that the proposed feature selection scheme is working well. You can also propose a novel classification framework for this problem. Subsequently, execute the best framework on the test data and submit the class labels in a text file. Each row of the text file will contain the serial number of each test data point followed by the class label separated by a comma e.g., d1, genre_type.
Skills: Machine Learning · Python (Programming Language) · Jupyter