**4.4.2 variation**

For different values the RSOM network presented a global accuracy significant variation. Table 5 shows the confusion matrices and figure 15 shows the superposition of the global accuracies due to the variation, for each RSOM dimension studied.

Figure 16 shows a ROC graph for the RSOM model with 5x5 grid and variation. One notices that with intermediate values of (=0.35 and =0.60) the classifier presented the greatest performances (larger tp rate and smaller fp rate).

Figure 17 shows a ROC graph for the RSOM model with 7x7 grid and variation. There is an approximation of the performances in this network, between the extreme values (=0.10 and =0.85) and the intermediate values of (=0.35 and =0.60). Despite this fact, the intermediate values still presented a general tendency to be the best performers, with larger tp rate and smaller fp rate.



Table 5. variation and global accuracy of the RSOM network in different grids

For different values the RSOM network presented a global accuracy significant variation. Table 5 shows the confusion matrices and figure 15 shows the superposition of the global

Figure 16 shows a ROC graph for the RSOM model with 5x5 grid and variation. One notices that with intermediate values of (=0.35 and =0.60) the classifier presented the

Figure 17 shows a ROC graph for the RSOM model with 7x7 grid and variation. There is an approximation of the performances in this network, between the extreme values (=0.10 and =0.85) and the intermediate values of (=0.35 and =0.60). Despite this fact, the intermediate values still presented a general tendency to be the best performers, with

Table 5. variation and global accuracy of the RSOM network in different grids

accuracies due to the variation, for each RSOM dimension studied.

greatest performances (larger tp rate and smaller fp rate).

**4.4.2 variation** 

larger tp rate and smaller fp rate.

Fig. 16. ROC graph for the RSOM model in 5x5 grid with variation

Fig. 17. ROC graph for the RSOM model in 7x7 grid with variation

Figure 18 shows a ROC graph for the RSOM model with 9x9 grid and variation. This figure indicates a superior performance of this classifier, when one uses intermediate values of (=0.35 and =0.60). In such cases the classifier becomes nearly ideal, with tp rate approaching 100% and fp rate near 0%. On the other hand, this classifier performance becomes very poor for the lowest extreme (=0.10) when compared to the other parameters.

Recurrent Self-Organizing Map for Severe Weather Patterns Recognition 173

 The U-matrix of the RSOM network presented a better cluster visualization (in terms of separation) when compared with other networks studied, allowing a clear

The labeling of the neurons in the maps, after the training, was better defined for the

 Finally, after the ROC analysis, it was concluded that among the neural networks studied, the RSOM had the best performance as classifier, confirming its usefulness as a

Angelovič, P. (2005). Time series prediction using RSOM and local models, *Proceedings of IIT.* 

Chappell, G. & Taylor, J. (1993). The temporal Kohonen map. *Neural Networks*, Vol.6, No.3,

Cherif, A.;Cardot, H. & Bone, R. (2011). SOM time series clustering and prediction with

Cooray, V.; Cooray, C. & Andrews, C. (2007). Lightning caused injuries in humans. *Journal of* 

Fawcett, T. (2006). An introduction to ROC analysis. *Pattern Recognition Letters*, Vol.27, Issue 8, (June 2006), pp. 861-874, ISSN 0167-8655, doi:10.1016/j.patrec.2005.10.010 Hagenbuchner, M.; Sperduti, A. & Tsoi, A. (2003). A self-organizing map for adaptive

Hammer, B. ; Micheli, A. ; Sperduti, A. & Strickert, M. (2004). Recursive self-organizing

Han, J. & Kamber, M. (2006). *Data Mining: Concepts and Techniques*, 2nd ed., Morgan Kaufmann Publishers, San Francisco, USA, ISBN: 978-1-55860-901-3 Huang, S.-C. & Wu, T.-K. (2010). Integrating recurrent SOM with wavelet-based kernel partial

Kohonen, T. (1990). The self-organizing map. *Proceedings of the IEEE*, Vol.78, No.9,

Kohonen, T. (2001). *Self-organizing maps*, Springer Series in Information Sciences, ISBN 3-540-

Koskela, T. ; Varsta, M. ; Heikkonen, J. & Kaski, K. (1998a). Recurrent SOM with local linear

Koskela, T.; Varsta, M.; Heikkonen, J. & Kaski, K. (1998b). Time series prediction using

*Intelligent Engineering Systems*, Vol.2, pp. 60–68, ISSN: 1327-2314

recurrent neural networks. *Neurocomputing*, Vol.74, Issue 11, (May 2011), pp. 1936-

*Electrostatics*, Vol.65, Issues 5-6, (May 2007), pp. 386-394, ISSN 0304-3886,

processing of structured data. *IEEE Transactions on Neural Networks*, Vol.14, No.3,

network models. *Neural Networks*, Vol. 17, Issues 8-9, New Developments in Self-Organizing Systems (October-November 2004), pp. 1061-1085, ISSN 0893-6080, doi:

least square regressions for financial forecasting. *Expert Systems with Applications*, Vol.37, Issue 8, (August 2010), pp. 5698-5705, ISSN 0957-4174, doi: 10.1016/j.eswa.2010.02.040 Jayaratne, R. (2008). Thunderstorm electrification mechanisms, In: *The Lightning Flash,* V.

Cooray, (Ed.), pp. 17-44, Institution of Engineering and Technology,ISBN 978-0-

models in time series prediction, *Proceedings of Sixth European Symposium on* 

recurrent SOM with local linear models, *International Journal of Knowledge-based and* 

potential tool for studies related to the severe weather patterns recognition.

RSOM network when compared with the other networks studied;

1944, ISSN 0925-2312, doi: 10.1016/j.neucom.2010.11.026

differentiation among the three clusters;

*SRC 2005,* pp. 27–34, April 27, 2005

pp. 441-445, ISSN: 0893-6080

doi:10.1016/j.elstat.2006.09.016

pp.491–505, ISSN: 1045-9227

10.1016/j.neunet.2004.06.009

67921-9, Berlin, Germany

85296-780-5, London, United Kingdom

*Artificial Neural Networks*, pp. 167–172

(September 1990), pp. 1464-1480, ISSN 0018-9219

**6. References** 

Fig. 18. ROC graph for the RSOM model in 9x9 grid with variation

In summary, we can say that the RSOM network was the one which offered the best clusters classification performance, as long as, using intermediate values.
