This paper presents a methodology to predict the geographical locations of Macedonian Tombs based on archaeological and geographical facts, geospatial analysis, predictive modeling and fuzzy logic. The methodology was developed upon the assumption that the location of those tombs was not randomly chosen, but it was a result of logical decisions based on geographical considerations and factors related to human activity (settlements, roads). To create a predictive model for this case, we had to do an extended archaeological bibliographical research on the topic, study the geographical data of the region of interest and collect the coordinates of the known tomb locations. The Barrington Atlas was the basis for the geographical data, apart from the digital elevation model, which was taken from NASA’s SRTM and was modified according to Barrington’s geological information regarding the ancient landscape. The idea was to create a model that would be able to provide map regions assigned with specified probability of site occurrence. To this end, six criteria were selected: elevation, slope, soil hardness, distance from settlements, distance from rivers and distance from roads. These criteria were transformed into normalized maps (GIS layers), which could be parameterized (by adjusting a buffer zone, b, around the median of criteria values) and were fed to the prediction model in order to provide probability maps. The prediction model consists of the weighted linear combination of the criteria:

S = Σ w_i x_i 

It should be noted that the weights are only adjusted for the criteria distance from settlements and distance from roads (based on the archaeological research) and the rest of the weights are equal and are adjusted accordingly so that the total summation of the weights equals one (1). By adjusting b and w, we have experimented on various combinations. For the validation of the experimental results (and the accuracy of the model), we have used the Kvamme’s predictive gain G, which is defined as:

G = 1 – E₁/E₂

with E₁ the % of total area where tombs are predicted, and E₂ the % of observed tombs within area where they are predicted, for a specified probability of site occurrence.

The experimental results for the case of high and very high probability regions (p>60%), where more than 75% of the known tombs were identified, show that the highest G value is achieved for w=2 and b=1.5. The prediction results are among the best attained in similar studies and validate the model as accurate and efficient to provide answers to a series of questions with reference to the problem at hand (archaeological research, cultural resource management and protection, land use, etc.).