| Title |
Didžiaisiais duomenimis apsimokančio algoritmo ir posistemio neįprastam laivų eismui atpažinti jūrų uoste kūrimas |
| Translation of Title |
Creation of algorithm for machine learning on big data and development of subsystem for recognition of abnormal maritime traffic in seaport. |
| Authors |
Venskus, Julius |
| Full Text |
|
| Pages |
62 |
| Keywords [eng] |
Big data ; maritime traffic ; abnormal traffic ; SOM ; virtual pheromone |
| Abstract [eng] |
At present, methods and algorithms that have been developed to recognise abnormal maritime traffic are not suitable for work with big data. For this reason, the learning of normal traffic pattern takes too much time. If smaller data is used, the precision of classification suffers, therefore algorithms for recognition of abnormal maritime traffic cannot reduce the workload for traffic control operators. The paper presents both an algorithm for machine learning on big data subsystem and subsystem which is based on modified Kohonen SOM network. SOM network was modified by adding a virtual pheromone tag to neurons thus creating a distance calculation function between neurons in non-Euclidean space. The Mexican hat wavelet was used for neural neighbourhood function. Direct machine learning with a fusion of two SOM networks was used for adaptation of SOM for learning from big data. A subsystem for recognition of abnormal maritime traffic was created on basis of the created algorithm. The subsystem integrates into maritime traffic and monitoring system. A prototype was created for verification of the subsystem and algorithm. The prototype verification and field test for algorithm was carried out using the big data of Klaipeda seaport maritime traffic. The classification precision of 90% at 80% sensitivity of classification was achieved. |
| Dissertation Institution |
Klaipėdos universitetas. |
| Type |
Master thesis |
| Language |
Lithuanian |
| Publication date |
2016 |
