Forecasting movie ratings with recommender systems

In this paper I construct different recommender systems to predict movie ratings and compare there performances. Recommender systems can be used to classify movie ratings for users as interesting or not worth watching and with a few simple statistics the classification performance can be compared for different models. The relevance of this topic is growing every year with the exponential increase in data available. A lot of study has already been done about this topic but there is still work to be done in increasing the forecasting power of different recommender systems. The most important recommender system in this topic is a neural network. Different models are added to compare the networks performance or try to improve it. The network uses singular value decomposition to extract less dimensional information for a large and sparse data matrix. I conclude that all models constructed in this paper do have a good in-sample fit but the out-of-sample classification power is relatively modest. With many ratings near the classification threshold out-of-sample classifying is harder.