README.md

🎵 Spotify Song Preference Dataset

This dataset contains Spotify audio features for 195 songs categorized as liked or disliked by the user. It was created to build and train ML models that can predict user preferences in music based on quantitative audio features.

---

📥 Dataset Overview

• Total songs: 195
• Format: CSV (data.csv)
• Source: Spotify API
• Target column: liked (1 = liked, 0 = disliked)
• Data type: Tabular
• Licensing: For academic and personal research use (derived from Spotify API)

---

📦 Dataset Composition

Category	Count	Description
Liked	100	Mostly French/American rap, rock, electro
Disliked	95	25 metal, 25 classical, 25 disco, 20 rap
Neutral (Pop)	❌ Not included (user is neutral)

---

🧪 Features

Extracted via Spotify API – "Get Audio Features for Several Tracks"

Feature	Description
danceability	How suitable the track is for dancing (0–1)
energy	Perceived intensity (0–1)
key	Musical key (0 = C, 1 = C♯/D♭...)
loudness	Overall volume in dB (-60 to 0)
mode	1 = major, 0 = minor
speechiness	Detects presence of speech (0–1)
acousticness	Confidence measure of being acoustic (0–1)
instrumentalness	Predicts presence of vocals (0–1)
liveness	Live audience presence (0–1)
valence	Positiveness of the song (0–1)
tempo	Beats per minute (BPM)
duration_ms	Duration of the song in milliseconds
time_signature	Estimated time signature (e.g. 4 = 4/4)
liked (target)	1 = liked, 0 = disliked

---

🔍 Exploratory Data Analysis (EDA)

✅ 1. Missing Values

• No missing values found

✅ 2. Class Distribution

• Liked (1): 100 songs
• Disliked (0): 95 songs
• Class is balanced

✅ 3. Data Types

• All features are numerical
• Target (liked) is binary

✅ 4. Summary Statistics

• Energy, Danceability, Valence tend to be higher for liked songs
• Acousticness and Instrumentalness higher in disliked songs

✅ 5. Correlation Matrix

• Strong positive correlation: energy ↔ loudness
• Negative correlation: acousticness ↔ energy, valence

✅ 6. Visual Highlights (Suggested)

• Boxplots: energy, danceability by liked
• Countplot: class balance of liked
• Heatmap: correlation of features
• Scatter: energy vs valence colored by liked

---

🤖 ML Use Cases

You can use this dataset to train:

• Logistic Regression
• Random Forest
• KNN / SVM
• ANN / XGBoost / LightGBM
• Naive Bayes

---

📊 Visualizations

1. Boxplot: Energy Distribution by Liked

This shows how energy values are distributed for liked and disliked songs.

---

2. Boxplot: Danceability Distribution by Liked

This shows how danceability varies between liked and disliked songs.

---

3. Scatter Plot: Energy vs Valence

This plot helps visualize clusters or spread of liked vs disliked songs based on energy and valence.

---

4. Correlation Heatmap

This heatmap shows how all audio features correlate with each other.

5. Countplot: Liked vs Disliked Songs

This chart shows the number of songs in each class: 0 = Disliked, 1 = Liked.
It confirms that the dataset is nearly balanced.

📉 Statistical Testing

To determine which features are statistically different between liked and disliked songs, a two-sample t-test was performed using:

Feature	p-value	Significance
danceability	0.0000	✅ Significant
energy	0.0159	✅ Significant
key	0.5371	❌ Not significant
loudness	0.0000	✅ Significant
mode	0.7418	❌ Not significant
speechiness	0.0000	✅ Significant
acousticness	0.0134	✅ Significant
instrumentalness	0.0000	✅ Significant
liveness	0.8924	❌ Not significant
valence	0.0002	✅ Significant
tempo	0.0000	✅ Significant
duration_ms	0.0000	✅ Significant
time_signature	0.0023	✅ Significant

🔍 Observation: - Features with p-value > 0.05 are statistically insignificant

- These features do not show a meaningful difference between liked and disliked songs

- We can safely remove the following features:
    liveness
    mode
    key

✅ This simplifies the dataset and improves model performance by removing noise.

📈 Model Accuracy Comparison

Bar chart showing accuracy of different models used in the project.