Spotify Recommendation System

Overview

Built a personalized music recommendation system that suggests songs based on listening history and audio feature analysis. The system combines collaborative filtering with content-based approaches for hybrid recommendations.

Problem Statement

Generic music recommendations often miss personal preferences. This project creates a personalized recommender that:

• •Learns from individual listening patterns
• •Uses audio features for content-based similarity
• •Provides explainable recommendations

Data

• •Source: Provided as a part of ML-I assesment
• •Features: 22 audio features per track
• •Size: 3600 unique tracks

Approach

1. •

   Feature Engineering

   • •Normalized and scaled features
   • •Introduced derived features
2. •

   Model

   • •One-Class SVM Configuration
3. •

   Evaluation

   • •System Recall for each User
   • •Average System Recall: 86%

Results & Impact

• •Average System Recall: 86%

Key Learnings

• •Hybrid approaches outperform single-method recommenders
• •Feature correlation analysis reveals interesting musical patterns

Links

• •GitHub Repository