Data-Driven Analysis of Music Valence Factors

Objective

• Analyze how each dimensions of a piece of music (key, mode, dancability, acousticness, etc.) influences the valencity (perceived emotional positivity) of the music
• Build a model to predict valence
• Determine which audio dimensions most strongly influence valence

Dataset

• Dataset obtained from kaggle： https://www.kaggle.com/datasets/vicsuperman/prediction-of-music-genre
• Cleaned and preprocessed the raw/dirty dataset for effective analysis

Methods

• Applied K‑means clustering (using an elbow curve to select k) to uncover natural groupings
• Visualized feature relationships with matplotlib charts and heatmaps
• Trained a Random Forest classifier to predict valence from audio features
• Extracted and ranked feature importances to pinpoint key drivers

Results

• Developed a model achieving 89.5% accuracy in valence prediction
• Identified danceability, energy, and acousticness as the top three contributors to emotional positivity