The AUC is not the only way to evaluate a model. The first question in evaluating the model is whether the model at least covers all known presence records. Another thing is that although the model has a high AUC, it is possible that the AUC deviation between the repetitions in the evaluation is high. A high deviation in AUC values during the evaluation could be caused by the presence of outliers. Additionally, outliers can create the model to be extended by sites that are not part of the optimum for the species. In a species of wide distribution, in which all the points of presence are occupied for the model can contain this type of data. If you mantain or remove that kind of data is a question of expertise. Another fact is that in species of wide distribution the model can be extended by habitats that are not adequate according to the knowledge of the expert on the species and that can only be determined when making the cut of the models with the presence threshold when looking for the realized niche. Be careful with bootstrapping when making models with more than 20 points of presence, this is because if a high number of repetitions is not used it is possible that not all points of presence are used in the evaluation of the model. In cases where there are more than 20 points of presence it is better to use cross-validation. In the other hand, if you never get a high AUC it is possible that the variables that you used to do the model are not the variables that determine the distribution of your species. In that case you have to search new variables beyond climatic, if it is the case. I hope it is useful. Greetings.

Higher AUCs are produced in Maxent when we are working with specialist species because of it does not incorporate true absences. However, an error is enclosed within this models. Really, in these cases, we should use to validate (absences) nearby points over distant points. Some ideas: you should set a coherent extent-area; you try to set different background numbers; you try to validate by using resampling methods.

References: 

Jiménez-Valverde, A., Lobo, J. M., & Hortal, J. (2008). Not as good as they seem: the importance of concepts in species distribution modelling. Diversity and Distributions, 14(6), 885–890.

Tsoar, A., Allouche, O., Steinitz, O., Rotem, D., & Kadmon, R. (2007). A comparative evaluation of presence-only methods for modelling species distribution. Diversity and Distributions, 13(4), 397–405. 

Depending how you define "ubiquitous"/ "widespread distribution" my experience seem to differ from yours. Endemic plants indeed mostly show high AUCs (for exceptions and their explanation see attached), but widespread bears show equally high AUCs in the same area. 

Have you used a fundamental set of environmental variables/predictors/co-variates? If you used categorical variables, did you optimise the number of categories?

Article Fine resolution distribution modelling of endemics in Majell...

Article Where the bears roam in Majella National Park, Italy

I wouldn't necessarily hang your hat solely on improving the AUC score.  Perhaps try running replicates or something like a 10-fold cross-validation and look at omission rates of your testing data (the held-out presence records).  

http://link.springer.com/article/10.1007/s10531-013-0606-1

Because common species are usually more generalist than the rarest ones, they are inherently difficult to model as their ecological requirements are less clear. Some species can be present nearly anywhere, so it can be very difficult to predict their distribution accurately. As AUC is sensitive to prevalence, increasing the number of absences would probably artificially increase the AUC value, but it doesn't necessarily mean that your model will be more useful. I had the same problem with some generalist birds like the common cuckoo, for which the best AUC that I obtained was 0.707.

First of all, I would use estimators other than the AUC, or use them in combination with the AUC (Warren et al., suggest AIC, others recommend TSS. I would personally use as many lines of evidence as possible as there is no agreement on which measure is the best -> AUC, AIC, TSS and omission rates will give you a good idea of your model performance when used together).

I would also follow Hein's suggestion of optimising the number of categories, as well as optimising the number and type of predictors. As you probably know slightly different sets of predictors might yield highly different AUC scores and predicted models. I found Hein's hierarchical method of predictor selection very useful to understand how different predictors affected my models 

I recommend AIC to evaluate different hypothesis in order to simplify the model. Maxent produces overly complex models. It is very important optimising the number of categories, but not only the number, we need to choose good options to integrate between among categories because of these new categories will determine the empirical probability of the occurences. Then, according to Jiménez-Valverde (2014):  “Sensivity or Specifity derived from the application of the threshold that makes them equal is a preferred measure of discrimination power. Together with the rate of false positives and negatives, and with the prevalence of the species, these statistics provide more information about the discrimination capacity of the models than the AUC”. On the other hand, the TSS is like a improved version of the Kappa index. 

References:

Allouche, O., Tsoar, A., & Kadmon, R. (2006). Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). Journal of Applied Ecology, 43(6), 1223–1232.

Halvorsen, R., Mazzoni, S., Dirksen, J. W., Næsset, E., Gobakken, T., & Ohlson, M. (2016). How important are choice of model selection method and spatial autocorrelation of presence data for distribution modelling by MaxEnt? Ecological Modelling, 328, 108–118.

Jiménez-Valverde, A. (2014). Threshold-dependence as a desirable attribute for discrimination assessment: Implications for the evaluation of species distribution models. Biodiversity and Conservation, 23(2), 369–385. 

The AUC is not the only way to evaluate a model. The first question in evaluating the model is whether the model at least covers all known presence records. Another thing is that although the model has a high AUC, it is possible that the AUC deviation between the repetitions in the evaluation is high. A high deviation in AUC values during the evaluation could be caused by the presence of outliers. Additionally, outliers can create the model to be extended by sites that are not part of the optimum for the species. In a species of wide distribution, in which all the points of presence are occupied for the model can contain this type of data. If you mantain or remove that kind of data is a question of expertise. Another fact is that in species of wide distribution the model can be extended by habitats that are not adequate according to the knowledge of the expert on the species and that can only be determined when making the cut of the models with the presence threshold when looking for the realized niche. Be careful with bootstrapping when making models with more than 20 points of presence, this is because if a high number of repetitions is not used it is possible that not all points of presence are used in the evaluation of the model. In cases where there are more than 20 points of presence it is better to use cross-validation. In the other hand, if you never get a high AUC it is possible that the variables that you used to do the model are not the variables that determine the distribution of your species. In that case you have to search new variables beyond climatic, if it is the case. I hope it is useful. Greetings.

Thanks for recommendations.

For an example of the use of several evaluation criteria in parallel, see our two articles testing algorithms and model evaluation criteria (Duque-Lazo et al.). Broadly, our findings suggest that the choice of evaluation criteria does not necessarily result in the selection of an alternative model than basing the model selection on AUC alone. Further, the comparison between the papers suggest that when using the GLM and maxent algorithms, the selection of predictor variables may have the greater impact.

For the non-statistical method, can apply ground-truthing or field survey to validate the model.

can any body help to calculate Kappa and TSS value for maxent out put

Imran Bodlah

Attached here you find an excel sheet that will help you to calculate TSS, Kappa and False Negative Rate (FNR).