Week 7
- Andrew DeJonge, Miguel Millan
- Jun 26, 2017
- 2 min read
Goals: 1.) Finish extracting features from the IoTPOT Data set
2.) Find equations/publications for Hierarchical HMMs
3.) Gather information on initializing HMM using statistical processes
4.) Update Abstract/Create new abstracts
5.) Figure out how to reduce features on data sets without the distribution of states
6.) Think more about intent, and how we can model intent with HMMs
-if possible, learn how to find intent based on the data we already have
Monday: Presented our week 6 presentation at the group discussion. Discussed with Coach and Ali about our plans for the rest of the program. Looked into ways to initialize HMMs in Matlab. Finished extracting features from the IoTPOT data set.
Tuesday: Met with Coach over how to better combine data and use a tiered model of HMMs to represent many different types of attacks on IoT. Revised our abstract/worked on making new abstracts. Began working on a new method of combining features and standardizing our data with statistical methods.
Wednesday: Finished implementing new statistical method of combining and standardizing feature data in matlab with similar accuracy to the previous method. Began writing a new feature reduction to find the best features of a data set. Looked more into ways to initialize Hidden Markov Models without knowing the distribution of states of the network over time.
Thursday: Worked on our presentation for the weekly discussion, worked on training Hidden Markov Models in Matlab, both with the new combination method and training from the NSL-KDD, then testing on the Kyoto data set. Met with Coach, Aziz, and Ali over our presentation and new Abstracts for papers #2 and #3.
Friday: Wrote new algorithm of optimizing a set of features for Hidden Markov Models with the best accuracy, as well as for feature pruning. Tried using unsupervised learning on the NSL-KDD set to test on the non-training portion of the NSL-KDD set with poor results, and optimized the DDoS data set with the new "alpha" optimization algorithm.