Klasifikasi Tingkat Konsentrasi Belajar Berdasarkan Fitur Audio Musik Spotify Menggunakan Large Language Model
Year: 2026
Type: dissertation
Abstract: Background music has been widely used to support learning activities, but its effectiveness in enhancing concentration depends heavily on its audio characteristics. This study implements a learning concentration level classification system based on Spotify music audio features using a technique for serializing tabular data into text by comparing five approaches: Tabular ML XGBoost, Classical NLP TF-IDF + XGBoost, LLM Zero-Shot, LLM Few-Shot, and BERT Fine -Tuning, using a dataset of 32,833 songs with features such as energy, instrumentalness, acousticness, and valence as the basis for classifying the labels HIGH, MEDIUM, and LOW. Evaluation results show that BERT Fine-Tuning achieved the best performance with an accuracy of 98.88%, followed by ML Tabular XGBoost (98.84%), Classic NLP (83.72%), LLM Zero-Shot (76.63%), and LLM Few-Shot (57.77%), proving that Transformer-based contextual representations are more effective at capturing audio feature relationships compared to other approaches, and indicating that fine-tuning is a more suitable approach for classifying imbalanced audio data.
Author: Raihaningtyas, Ayulexa Rifka
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7219156587