We analyzed a decade of movies, developed a machine learning model, and turned it into a flask app to predict if a movie will be a hit or a flop. Just fill out the form with some basic details, such as your budget, expected running time, director, actors, even the plot. We can predict with 71% confidence if your proposed movie will be successful.
- Train a supervised model to predict ratings of proposed movies.
- Create an historical data analysis with ten years of movie data.
- Create an interactive web app with movie rating predictions and a dashboard of visualizations.
- Extract using an API calls to obtain list of movies by year from OMDb.
- Transform using Python/Pandas to merge dataframes, replace null values with ‘NA’ or ‘0’ as appropriate, remove duplicates, normalize all headers
- Load the final final output to CSV and our SQL database.
We analyzed the data in Python/Pandas with matplotlib, pyplot, seaborn, and wordcloud.
Some observations:
- Surprisingly, we discovered IMDb user votes and IMDb critical ratings aren't strongly correlated.
- Budgets and IMDb ratings aren't strongly correlated.
- Interestingly, budgets and box office gross amount is correlated.
- Ratings from Rotten Tomatoes and Metascore don't always reflect the same kind of feedback for the same movie.
- Create labels based on IMDb rating.
- Select beneficial columns for the model.
- Convert categorical data to numerical using One-Hot-Encoding.
- Additional cleanup of null and missing values.
- Reduce number of distributions of some features such as country, language, genre, director, writer, actor, star, and rating (i.e. PG vs. R, etc.).
- Binning the IMDb rating into 3 distinct groups of <5, between 5 and 7, and >7.
- Encode the rest of categorical data using
get_dummies. - Process
plotvalues into vectors. - Merge all features into a single dataframe.
Our final dataset uses the following features: runtime (mins), budget (USD), rated, director, writer, actors, language, country, genre, star, and plot.
We used a supervised machine learning classification model leveraging a Random Forest Classifier. The greatest challenge was in converting the plot data into a vector to join with the rest of the model. We used a multi-step process to clean and transform words into a meaningful representation, but we did it!
We used a Flask app to serve our website with a dashboard highlighting interesting relationships with features of the data set and our prediction app. We wrote a function to bring together the vectorized plot data with the model which produces the prediction.