The Simpsons Members:

• Cameron Carlin
• Bradley Kenstler
• Alice Zhao
• Melanie Palmer

Google Slides Presentation: https://docs.google.com/presentation/d/16pyf2bN2KBMHmWfQ47IwyxvP_73TRdx7vIDEKpzpCwU/edit?usp=sharing

About the Data:

Data Source: https://www.kaggle.com/wcukierski/the-simpsons-by-the-data

Data Description: This dataset contains the characters, locations, episode details, and script lines for approximately 600 Simpsons episodes, dating back to 1989. Contains approximately 131,000 individual lines of speech.

Problem we would like to solve with this data:

Using the lines of speech from every Simpsons episode ever, we aim to accurately classify which character is speaking based off what is spoken and where.

Methodology to solve this problem:

• Narrow down important locations, as particular characters tend to speak and hang out in particular areas.
• Criteria for word frequency and document frequency (via TFIDF)
• Stemming words to turn down the count of 'unique' words

Algorithms Attempted:

• Logistic Regression (L1 and L2)
• KNN-Classification
• Random forest
• SVM
• Decision Tree
• AdaBoost

Criteria to determine the best model:

• Tuning accuracy for ‘main’ characters versus others: there are over 6,000 characters in the full collection of lines, where it would be ideal to tune the model to the more common characters for both accuracy and interest.
• Misclassification rate, balance between true positive and false positive