Inspiration
We all enjoy listening to music, encompassing various kinds of music. There are popular songs and those that are not so popular. We know many songs that aren't popular but sound good. Hence, we want to explore what features contribute to a good song apart from the artist's popularity, and which songs possess these potential attributes
What it does
-It enables the music apps to recommend songs that possess specific features (excluding artist popularity), increasing the potential for those songs to become popular. -It provides a platform for record companies to discover artists with promising potential, ultimately contributing to the success of emerging talent.
How we built it
-We first analyze the changes of various genres in these 10 years, and show the popularity of genres in these 10 years through line graphs and word cloud graphs. Then we selected 15 elements of song as our predicted features by filtering the features, taking popularity as the dependent variable. We next fit machine learning models such as Random Forest and perform tuning.
Challenges we ran into
-Each of our accounts has a limit of 1000 API calls. It's challenging for us to obtain large amounts of complete data due to this restriction. We need to constantly debug, borrow accounts from friends, and then merge and clean the data. -Our time is limited, and the volume of data is vast, requiring a long processing time. Thus, we have very little time for debugging, and the accuracy of our code is of utmost importance.
Accomplishments that we're proud of
-We successfully learned to use machine learning models and algorithms, including the logistic model, random forest, SVM (Support Vector Machines), XGBoost, Naive Bayes, and neural networks. Within a limited timeframe, we managed to obtain a substantial amount of data to support our research. The accuracy and precision of our models are overall quite satisfactory. -We learned the value of teamwork and had excellent communication within the team. We've established profound friendships among u
What we learned
-We've learned the process of training machine learning models and successfully split the data into testing and training sets. -We've learned machine learning models such as the logistic model, random forest, SVM (Support Vector Machines), XGBoost, Naive Bayes, and neural networks. -We've learned to use GitHub as a team.
What's next for Song Hit Prediction Using Machine Learning
-We can delve deeper into the relationships between each feature, investigating if there are any missing features that could enhance our accuracy. Additionally, we can consider larger contextual factors such as societal influences or conduct time-series analysis on individual features.We can also analyze more data and use a Recurrent Neural Network (RNN) to develop a song recommendation algorithm.
Log in or sign up for Devpost to join the conversation.