In my thesis, I present the theoretical and practical operation of two algorithms to search for anomalies in time series, one of which, named HOT SAX, is the optimized version of the other basic one. These algorithms are suitable for data mining in time series without any model specifications and thus are useful in the specified task, which is evaluating the signals originating from, in some way, limited people’s daily routine.
First I present the theoretical workings of the mentioned algorithms, and every other information on this topic, which is necessary for understanding this operation. This includes the details of time series and one time series representation, named SAX, with which the above mentioned HOT SAX algorithm is able to work more easily and more quickly, than the original time series. Then, I describe the principles of the two algorithms operation and the differences between them, and this ends the theoretical part of the thesis.
The second part of my thesis contains the implementation of the described algorithms and the tests performed on these. This basically covers the description of the implementation steps and the presentation and the reasoning of the most important design decisions. After that I complete the thesis with describing the tests performed on the application, to show that the implementation is correct and to demonstrate the differences between the two algorithms.