Spatial Modelling of Interactions Between Dengue Incidences and Changing Climate by ..... SPATIAL MODELLING OF INTERACTIONS BETWEEN DENGUE INCIDENCES AND CHANGING CLIMATE BY INTEGRATING ANN TECHNIQUE WITH GIS

According to various studies, it is well established that climate characteristics are one of the significant factors influencing vector-borne diseases and their long-term variations in a climate change scenario may affect the vector-borne diseases. According to World Health Organisation (WHO) factsheets (2016), more than 2.5 billion people in over 100 countries are at risk of dengue alone which is one of the deadliest vector-borne diseases. Therefore, it is very vital to develop a surveillance system which is capable of predicting the high-risk areas so that the proactive and effective control measures can be taken immediately. In the present study, a spatial data mining model is developed by integrating artificial neural network (ANN) technique into a Geographic Information System (GIS). A statistical method such as logistic regression has been used to detect the areas where the prevalence of the disease is high. Also, possible associations between disease incidences and meteorological parameters have been investigated. This model will highlight areas which are at high risk of dengue by examining the interactions between dengue fever incidences and environment. The primary purpose of the present work is to provide a better understanding of the spatial dispersal of the dengue fever risk in the rural as well as the urban areas of Delhi. Also, this model will bring new insights to the public health officials and policymakers to reduce the risk of deaths mainly in rural areas due to lack of awareness and health facilities.


Introduction
The vector-borne diseases such as dengue is one of the most rapidly spreading mosquito-borne diseases. According to World Health Organisation (WHO), in the world, the estimated number of dengue occurrences are 50-100 million every year of which 15-30 million incidences happen in India (WHO factsheet 2013(WHO factsheet , 2014. Prevention and control of dengue fever (DF) is an effective method to reduce the mortality rate. GIS plays a vital role in disease susceptibility mapping to control the spread of the disease. The generated risk maps help to identify potential areas of DF due to their social and environmental conditions. These maps predict the distribution of DF incidences which help public health officials and planners to develop effective control strategies and thus enabling the establishment of early warning system that could contribute in minimisation of the DF occurrences due to the influence of various causative factors. According to various studies, meteorological parameters have a significant impact on DF incidences. This impact is due to parameters such as temperature and humidity which affect the life cycle, the rate at which mosquito bites, infectious and survival rates of mosquitoes and on the incubation period of dengue virus (Hii et al., 2009). As temperature increases, Aedes mosquitos' ,carrier of dengue virus displays shorter periods of growth in all stages of life-cycle causing an increase in the vector density. Xu et al. (2014)  According to various studies, susceptibility maps could be generated by modelling the relationship between DF incidences and essential physiographic and environmental factors (Tsegaw et al., 2013;Yang, 2016). There are two types of disease mapping techniques such as knowledge-driven and data-driven.The knowledge-driven approach is based on the expert's knowledge to assign weights to a set of factors whereas data-driven approach is based on the retrospective data patterns and relationships. The advantage of data-driven approach over knowledge-driven approach is that it provides a more straightforward and more direct method for disease susceptibility mapping.
Among the various data-driven methods, logistic regression (LR) and artificial neural networks (ANNs) are the two most widely used methods.

LR is a multivariate regression technique that is
Journal of Rural Development,Vol.37,No. (2), April-June:2018 used to predict the probability of presence or absence of a particular condition. This statistical method has been extensively used to generate risk maps. ANNs are widely used to manage multidimensional non-linear features of practical problems. However, very few studies have used these techniques in producing disease susceptibility maps.
In the present study logistic regression has been used to determine the association between meteorological parameters and DF incidences.
Also, an attempt has been made to implement radical basis functional network for producing disease susceptibility maps using Python and ArcGIS.

Methodology
In the present study, the methodology can be classified into two sections. Firstly, identification of an association between critical meteorological parameters and DF incidences. Logistic Regression: Regression can be defined as a process to determine a series of coefficients that describe the association between the independent variables and the dependent variables effectively. LR is a regression technique used to determine the relationship between several independent variables and the probability of a binary or categorical response (Lee and Sambath, 2006). The main advantage of LR is that when a suitable link function is added to a linear regression model then the variables can be any combination of continuous and discrete variables and they do not have to follow a normal distribution (Lee and Sambath, 2006). This regression technique could be used to predict the probability of disease incidences by considering various critical causative factors because the value of response variable is the   (Looney, 2002).

Results and Discussion
In the present study, multi-nomial logistic regression has been performed. The independent variables are minimum temperature, absolute humidity and mean wind speed. The dependent variable is the category of risk of DF incidences. .

Number of Dengue Cases Vs Wind Speed
Journal of Rural Development,Vol.37,No. (2), April-June:2018 established. In Table 1    In the present study, according to the results of logistic regression, it is evident that meteorological parameters are affecting the spatial as well as the temporal distribution of DF incidences. The impact of temperature has been associated with an increase in the DF incidences because of the increase in larval abundance which is in line with the studies performed (Pinto et al., 2011;Kumar et al., 2015). Also, by analysing the DF susceptibility maps, it can be observed This study is useful in determining the potential areas of DF occurrences particularly in rural areas where early warning of DF would help to take immediate actions to control its spread and also fast medical facilities could be provided by identifying the locations of DF occurrences.

Conclusion
The association of DF incidences with meteorological parameters in the study area Delhi has been analysed and the study period is