We will be predicting the scores of NFL games using team statistical averages from historic box scores. (Kaggle: Lawfty #DSCO17 Hack Night!) Dimension Reduction and Feature Selection Principal Component Analysis AIC/BIC Bayesian Lasso/Ridge Regression: Bayesian ridge regression is similar to ridge regression however it includes information about the features to determine the penalty weight.
Learning CART, Random Forest, GBM, SVM, Neural network
Ensemble of models: at the final stage we will try ensemble of different models to achieve better model performance
Model Evaluation: Randomly split the data into training set and test set (around 9:1). Predict the home score and away score for each test data and calculate the MAE for each prediction. Compare the MAE for each prediction model and choose the best one. Based on the models we get, we will also apply a ensemble learning method and check whether it will achieve a better prediction performance on our dataset.