Trained Neural Network Predicts Hollywood Blockbusters / Flops

Ramesh Sharda, an information scientist at Oklahoma State University in Stillwater, has trained an artificial neural network to recognise successful movies from flops with high degree of accuracy. Let’s look at the results.

The data consisted of 834 movies released between 1998 and 2002. The system ranked movies in 9 categories ranging from “flop” (total takings less than $1 million) to “blockbuster” (over $200 million).

The trained neural network was able to predict accurate category with 37% accuracy. However it could predict category correct within one category either side (+/- 1) with 75% accuracy. Link

He found the key factors identifying the ranking are:

• the “star value” of the cast
• the movie’s age rating
• the time of release against that of competitive movies
• the film’s genre
• the degree of special effects used
• whether it is a sequel or not
• the number of screens it is expected to open in

Significantly it doesn’t include the plot or author. What is interesting is that now blockbusters can be created on demand and you can continue to pay poorly the authors while spoiling the “stars”.

I think the experiment will be very easy to reproduce for Bollywood or Tollywood movies.