This thesis explores the possibility of predicting the mood a song will evoke in a person based on certain musical properties that the song exhibits. First, I introduce the topic of data mining and establish its significant relevance in this day and age. Next, I explore the several tasks that data mining can accomplish, and I identify classification and clustering as the two most relevant tasks for mood prediction based on musical properties of songs. Chapter 3 introduces in detail two specific classification techniques: Naive Bayes Classification and k-Nearest Neighbor Classification. Similarly, Chapter 4 introduces two specific clustering techniques: k-Means Clustering and k-Modes Clustering. Next, Chapter 5 implements these previously discussed classification and clustering techniques on a data set involving musical property combinations and mood, and makes conclusions about which musical property combinations will most likely evoke a happy, sad, calm, or agitated mood. It can be concluded very generally that songs that are in major keys, have consonant harmonies, faster tempos, and firm rhythms will most likely evoke a happy mood in a person. Songs that are in minor keys, have slower tempos, and firm rhythms will probably evoke a sad mood in a person. Songs that are in major keys and have flowing rhythms tend to evoke a calm mood in a person. And last but not least, songs that are in minor keys, have dissonant harmonies, and firm rhythms will evoke an agitated mood in a person.
Smith-Polderman, Sarah, "Let's Get in the Mood: An Exploration of Data Mining Techniques to Predict Mood Based on Musical Properties of Songs" (2013). Senior Independent Study Theses. Paper 4933.
Bachelor of Arts
Senior Independent Study Thesis Exemplar
© Copyright 2013 Sarah Smith-Polderman