A Support Vector Machine-Firefly Algorithm based forecasting model to determine malaria transmission Sudheer Ch a,n , S.K. Sohani a , Deepak Kumar a , Anushree Malik c , B.R. Chahar c , A.K. Nema c , B.K. Panigrahi b , R.C. Dhiman d a Department of Civil Engineering, Indian Institute of Technology, Hauzkhas, New Delhi, India b Department of Electrical Engineering, Indian Institute of Technology, Hauzkhas, New Delhi, India c Center for Rural Development and Technology, Indian Institute of Technology, Hauzkhas, New Delhi, India d National Institute of Malaria Research, ICMR New Delhi, India article info Article history: Received 16 April 2013 Received in revised form 31 August 2013 Accepted 20 September 2013 Communicated by Swagatam Das Available online 1 November 2013 Keywords: SVM Malarial incidences Forecasting Time series FFA abstract Accurate and reliable forecasts of malarial incidences are necessary for the health authorities to ensure the appropriate action for the control of the outbreak. In this study, a novel method based on coupling the Firefly Algorithm (FFA) and Support Vector Machines (SVM) has been proposed to forecast the malaria incidences. The performance of SVM models depends upon the appropriate choice of SVM parameters. In this study FFA has been employed for determining the parameters of SVM. The proposed SVM-FFA model has been adopted in predicting the malarial incidences in Jodhpur and Bikaner area where the malaria transmission is unstable. Monthly averages of rainfall, temperature, relative humidity and malarial incidences have been considered as input variables. Time series of monthly notifications of malaria cases has been obtained from primary health centers and from other local health facilities for a period of January 1998 to December 2002 in the region of Bikaner and from January 1998 to December 2000 in Jodhpur region. Further, the rainfall, relative humidity and temperature data have been obtained from meteorological records. The performance of the proposed SVM-FFA model has been compared with Artificial Neural Networks (ANN), Auto-Regressive Moving Average method and also with Support Vector Machine. The results indicate that the proposed SVM-FFA model provides more accurate forecasts compared to the other traditional techniques. Further, it has been recommended to carry out additional strides to explore the utility and efficacy of SVM-FFA model. Thus SVM-FFA can be an alternate tool to facilitate the control of vector borne diseases like malaria. & 2013 Elsevier B.V. All rights reserved. 1. Introduction Diseases caused by the infectious micro-organism which is transmitted to human beings by means of blood sucking anthropods are known as vector-borne disease. Among several anthropods, mosquitoes are the most common insects which are responsible for causing several diseases like malaria, dengue, etc. Among the vector born diseases, malaria is a great cause of fear for human health. Especially in the developing countries like African countries and India, thousands of people suffer from this disease every year. In spite of many studies on the malaria [1–3] still there were 243 million malaria cases reported in 2008 [4]. Malaria is still a major public health problem as 109 countries are declared endemic to the disease in 2008 [5]. Government authorities are incurring huge cost to control/eliminate the outbreaks of malaria. Due to global warming, rapid climate changes are occurring which result in the increase or decrease of malaria transmission depend- ing upon the specific micro-climate of that particular region. Temperature fluctuations affect the life cycle of vector as well as parasite [6]. Further, the process involved in malaria transmission is very dynamic and is completely site specific. Thus, it is great challenge for the researchers to predict the malarial outbreaks in advance. In the absence of knowledge about probabilistic attack of these diseases, government fails to provide adequate treatment facility on time. Thus, it is necessary to forecast the occurrence of these diseases in advance so that its devastating impact on the society can be reduced. There are various traditional techniques in forecasting time series models like Auto-Regressive (AR), Auto-Regressive Moving Average (ARMA) and Auto-Regressive Integrated Moving Average (ARIMA) models [7]. Such traditional methods do require mini- mum computational efforts to set up forecasting model which is considered to be an advantage. But, with non-linear nature of malaria occurrence it becomes difficult to use these models [8]. Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/neucom Neurocomputing 0925-2312/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.neucom.2013.09.030 n Corresponding author. Tel.: þ91 9999152196. E-mail address: sudheer108@gmail.com (Sudheer Ch). Neurocomputing 129 (2014) 279–288