Predictive Analytics for College Football -- 2

I would like to follow a CRISP-DM methodology 1.) understanding the domain and development goals of the Project 2.)identifying, accessing and understanding the relevant data source 3.)per-processing, cleaning, and transforming the relevant data 4.)developing models using comparable analytical techniques 5.)evaluating and assesing the validity and utility of the models, against each other goals of the study 6.) deploying the models for use in decision making ***all models will have numerical output variable. In the regression-based classification prediction, the numerical output model will be converted to a classification type by labeling the positive WinLoss numbers "WIN" and the negative WinLoss numbers "Loss" and then tabulating them in the confusion matrix. Excel data-set 2018-2011 K-fold cross validation for training/validation/testing subsets 3 separate teaching/validation testing data set's I.) NN model II.) Decision Tree III.)TBD ***if needed i.)Neural Network (MLP) w/back-propagation type supervised learning.***(I would like the NN model to run unsupervised if possible????) ii.)Decision Tree: Algo's : Information Gain, Gini index, Chi-square, used to identify a variable (from the pool of available variable's) and the corresponding threshold for that var in order to split the pool of observation into two or more [login to view URL] step is repeated at each leaf node until the entire tree has been constructed. (CART or C&RT) for classification and regression trees Second is becoming "Nostradamus Status"! I have an idea of what model will predict what results i am looking for this project. The larger question is applying the model to the betting line to increase predictability of beating bookmaker. Currently, the "odds", maybe by design have a win ratio of 50/50 or 49/51. The odds are noise though to AI models, I want the finished model to be able to predict accuracy through training/validation/testing enough confidence to choose the value of the line from it's own model, not the mere Bookmakers suggestion. So introduce Mr. football from above to predict scores and line totals to identify weekly match ups for 2019 . Sound fun? 1.) clean and prep databases for agreed upon testing models 2.) run (2) separate models on same data set and compare Results 3.) Include or exclude model results and back test on 2018 season games. All testings results need to be logged and preserved. (before/after) All validation iterations are to be logged and preserved incrementally All Learning iterations are to be logged and preserved... The finished model will have exclusive rights of ownership and proprietary results to James R Mahassek. This model, testing patterns and results will not be shared, given, borrowed or sold for any reason without my permission. ^^I will grant you permission exclusively to use the program for your own personal gain to be used for it’s intended purpose, as long it is not discussed or promoted. A confidentially agreement will need to be signed before we begin. ^^TBD It is going to be a lot of data 8 years with multi inputs each year. So first I will provide you with databases. You scope out the project cost and time, and we can get started.