Machine Learning for 20 time series datasets

I am a visualization scientist. I am looking for help from data mining/machine learning experts with a study for analyzing and predicting the eyewall replacement cycles (ERC) in hurricane simulations though time series data. Data description: 20 datasets, each has a total of 29 interested variables of 126 hour forecast at 6 or 3 hour interval from two model output files, ATCF track and SHIPS. The 20 track files and 20 SHIPS files are produced by 20 cycles of model output. Each trackfile has 43 data points from time steps 0, 3, 6, …, 126, and each data point has 17 selected variables. Each SHIPS file contains 22 data points from time steps, 0, 6, 12, …, 126, and each data point has 12 selected variables. (Please read the data sheet attached. Find the track and SHIP files to be downloaded from the clickable links in the datasheet with selected variables ) Tasks: Do both or either of the two below (1) Develop a method to predict whether there will be ERC or estimate the possibility of ERC in next 6, 12, and 24 hours for a given hour at a given test time series dataset. (2) Uncover the causal relationship between ERC and the given variables or their changes in last 6, 12 or 24 hours. Delivery: (1) a PPT file (telling why and what method is used, your findings, e. g. which variables are possibly linked with ERCs and any patterns/trends happen before ERCs occur, and visualizations/images), (2) all or key parts of your source code (so your results can be reproduced by me when needed) Time requirement: I need the job to be done as soon as possible, better by this Friday (4/13) in order to make a report due on Saturday (4/14). So, the code can be delivered later than the PPT.