 Ladies and gentlemen, we would like to thank the organizers for the opportunity to present to you our paper about modeling cultural change. In this talk we would like to focus on two issues. First, with the use of predictive modeling and particularly fuzzy logic and neural network, we would like to explain a spatial distribution of the sites and secondly using spatial temporal modeling we want to investigate possible interactions among sites. So to speak about chronological and geographical background we are focusing on the period of middle neolithic which was from 5300 to 4900 BC and in this time period there were spread for archaeological cultures. From the geographical background our study area consists of river Ipele with all odds of its tributaries. From this area we collected by study of the literature and by field walking around 500 components. To the predictive modeling were used only those components which were sufficiently dated and localized in space. I will present you model creation by fuzzy logic and my colleague will present you neural networks. So in both of our approaches we used these seven environmental parameters. In each of the parameter there is a unique distribution of values. In the fuzzy logic which is logic very suitable for archaeology because it explains very well uncertainty in the data there is a necessity for four control points in order to transform the input data in a scale from zero to one. So in order to have some objective measure how to take these control points a series of person tiles was used. We used Saga GIS software and this software offers you two tools. One tool is Fuzzify which transforms your data into the scale from zero to one and another tool is called Fuzzy Intersection which combines the transformed layers into the predictive model. There are used three different variants of the final predictive model calculation and we compared all of them. The explanation of the procedure will be based on the sites of LBK and Geli Azotse group. Here in this first chart you can see the values of Gini coefficient. Gini coefficient is a measure of distribution. Values close to zero mean perfect equality and values closer to one means perfect inequality. In order to better explain this chart we calculated number of components in the zone with high probability and with the zone with low probability and as we can see the Gini coefficient corresponds very well with zone of low probability of sites. Therefore for our purposes we should take a closer look to the models based on the fifth person tile but we can compare two extreme cases model based on fifth person tile and model based on 50th person tile. In order to validate these two models we used so-called internal validation which is based on an assumption that our predictive model is validated if the majority of the site is located in a zone with higher potential and this zone should have the least extent. So if we take a look at this line of the tables we see that the model based on fifth person tile is validated through internal testing that means we used the same sites as the model was calculated. So here is a visualization of the predictive model and this model was tested externally. We took a new set of test sites which were discovered very recently by Phil Volking within the ESAP project. Here in this table we can see that the spatial attributes of these new test sites resemble quite a lot the original data set so we assume that these sites will fit very well within the existing predictive model. So we extracted the cell values within the point locations and we see that most of the sites fall within the model based on the fifth person tile so again we see that this model is working and it's also validated through external set of the site. Last step was creation of the final predictive model which consists of both original set of sites and the test sites and we see through internal testing that this model works even better. It explains 76% of the site and the zone with highest potential has approximately 35% of the extent of the total area. Regarding the contemporary cultures which are eastern LBK and Buk culture with the same methodology with the same procedure we were not successful because almost the the majority of the site fall within the zone of low potential so the models are not validated and we are thinking about two reasons probably we were working with a low sample size you can see we were working only with the nine sites or there could be some other probably non-practical reasons which somehow affected the distribution of the sites in the landscape. Now I would like to ask my colleague Niklas to talk about neural networks. We would like to present some preliminary results of modeling using machine learning mainly because of lack of access to sufficient hardware this must be regarded as a work in progress but we believe it might be of interest to you and we'd love to get some feedback on it. The training area for the model is a nearly 2000 square kilometers area in the central Epoi River drainage basin. In the area are 64 known middle Neolithic sites and for this we used GRAS GIS and R with the RFRBS fuzzy rule based system package. The modeling was made with HIFAS hybrid neural fuzzy inference system system based on neural networks. Its fuzzy rules use the Mandani model and uses Gaussian membership function. The training data of this the area was of a 25 meter resolution and it was decreased by random sampling to about 15 percent so this was added the sites with a 100 meter buffer. The training data consisted of seven variables as we have used before with one exception instead of local elevation we had used landform which was created in the GRAS GIS module R point geomorphon which calculates terrain forms and associated geometry using machine vision approach and is represented by an integer from 1 to 10 from flat area 1 through slope 5 to depression 10. The model created by machine learning was then tested on smaller areas in the Middle Epoi River area. One 240 square kilometer area was tested at 100 meter resolution and a smaller about five square kilometer area at 25 meter resolution. Now unfortunately we cannot give you information on statistical significance however we believe a visual evaluation indicates quite plausible and promising results. This was the case for the 25 meter data as well as the 100 meter resolution data. Obviously the 25 meter data produced a lot more barrier than final result so we we now feel comfortable to continue with this work to test bigger areas to do statistical evaluation of the results and elaborate on the input rivals. So let's move to the second question and it was how we can model the relation between sites. To answer this question we take a look at the spatial temporal modeling which consists of a principle charted here in the upper part so the relative chronological dating of archaeological sites is transformed to absolute chronological time scale. To each site there is a certain some temporal uncertainty and you can visualize your results very much. With this approach you can use all of your data even uncertain data and you can take a look at the the phenomenons from spatial temporal aspect and you can compare compare your results. One of the cons we could think is that this approach is highly dependent on the input parameters. So we can now speak particularly in our data set. Here you can see our categories of dating and their temporal uncertainty and this temporal uncertainty was then transformed into absolute chronological time scale. You can see that the development of the Neolithic development in the river Ipari was quite stable. After the early Neolithic we see high increase of the sites, high increase of the components and these components gradually decreased towards the end of the Neolithic. So let's see a graphical visualization of this temporal uncertainty in the beginning of the Neolithic. You can see that in different parts of our studied area settlement cores were created. In the next time block we see a high increase of the components and the settlement areas expanded. This is also a time period of Gelezoce group and big culture and as we can see there is no culture or boundary besides they're communicating with each other. In the next time block we see that the sites, the number of components did not change. The density of the components in the eastern part remained the same but something happens in the western part and we see that in the next time block there are created two large group of the sites and in the eastern part we see some increase in the density which confirms itself in the next time block we see here a low density of the settlements and quite a lot of settlement clusters in western part of the studied area. So to sum up predictive model for the LBK and Gelezoce group was quite successful. It explained almost 80% of the sites in the landscape. These sites were located in on fertile soils soils in warm climate so we think about the orientation towards agricultural way of life. The contemporary eastern LBK and big culture the predictive model was not very successful and probably sites were located in a colder climate not so fertile soils and we can think about probably non practical reasons. With the use of spatial temporal modeling we can see that there is no cultural boundary between distinct cultural groups and this is more of an issue of cultural historical paradigm. In the future we would like to further test our predictive models and we would like to focus on neural networks. Last but not least we want to thank Silvia Fabian, Silvia Guba and the ISSA project for the possibility to work with unpublished data and we would like to thank you for your attention.