Project README: Spotify Hit Predictor For access to the colab, DM me.
Dataset Overview:
The dataset to be analyzed is "Spotify_dataset.csv," containing information regarding music tracks. This dataset is sourced from Kaggle and is primarily derived from the Hit Predictor dataset. It comprises over 40,000 tracks labeled as hits or flops, along with their features. The dataset aims to support machine learning tasks related to predicting track success and characterizing music tracks. In this notebook, I will aim, in addition to data preprocessing the feature pruning, to undertake the following tasks:
Task 1: Supervised Learning
Objective: Predicting whether a track is a hit or a flop.
Secondary Objective: Predicting discrete valence.
Task 2: Unsupervised Learning
Objective: Characterizing music tracks.
Dataset Information:
Name: Spotify Hit Predictor Dataset (1960-2019)
Content:
Features for tracks fetched using Spotify's Web API.
Tracks labeled '1' or '0' (Hit or Flop) based on certain criteria.
Purpose: This dataset enables the creation of classification models predicting a track's success.
Project Resources:
Dataset: "Spotify_dataset.csv" provided alongside this project description.
Additional Information:
The dataset from which this one was created: https://www.kaggle.com/datasets/theoverman/the-spotify-hit-predictor-dataset
Models Applied:
Several models were applied, including:
Multinomial logistic regression
Tree Model
Multi-Layer Perceptron
Bernoulli Bayes
K-nearest neighbor classifier distance model
These models were selected based on comparisons with other models of each type, such comparisons are found in the code, (E.G.: Comparing Gaussian and Bernoulli Naive Bayes)