Developing a dengue forecast model using Long Short Term Memory neural networks method

Abstract

22Background 23 Dengue Fever (DF) is a tropical mosquito-borne disease that threatens public health 24 and causes enormous economic burdens worldwide. In China, DF expanded from 25 coastal region to inner land, and the incidence sharply increased in the last few years. 26In this study, we conduct the analysis of dengue using the Long Short Term Memory 27 (LSTM) recurrent neural networks. This is an artificial intelligence technology, to 28 develop a precise dengue forecast model. 29 Methodology/Principal Findings 30The model is developed from monthly dengue cases and local meteorological data of 31 2005-2018 among top 20 Chinese cities with a record of the highest dengue 32 incidence. The first 13 year data were used to construct the LSTM and to predict the 33 dengue outbreaks in 2018. The results are compared with the estimated dengue cases 34 of other previously published models. Model performance and prediction accuracy 35 were assessed using Root Mean Square Error (RMSE). With the LSTM method, the 36 prediction measurements of average RMSE drop by 54.79% and 34.76% as compared 37 with the Susceptible Infected Recovered (SIR) model and Zero Inflated Generalized 38 Additive Model (ZIGAM). Our results showed that if only local data were used to 39 develop forecast models, the LSTM neural networks would fail to capture the 40 transmission characteristics of dengue virus in areas with fewer dengue cases. 41 Contrarily, transfer learning (TL) can improve the accuracy of prediction of the 42 LSTM neural network model in areas with fewer dengue incidences. 3 43 Conclusion and significance 44 The LSTM model is beneficial in predicting dengue incidence as compared with other 45 previously published forecasting models. The findings provide a more precise forecast 46 dengue model, which can help the local government and health-related departments 47 respond early to dengue epidemics. 48 49 Author summary 50In China, DF is a public health concern that poses a great economic burden on local 51 governments. However, the incidence has sharply increased in recent years with 52 growth in the sub-regions. With this issue, it will be challenging to develop an 53 accurate and timely dengue forecast model. LSTM recurrent neural networks, deep 54 learning methods and virus propagation rules by learning from observational data 55 offer more advantages in predicting the prevalence of infectious disease dynamics 56 than the traditional statistical model. The 2005-2017 data of the top 20 Chinese cities 57 with the highest dengue incidence were used to construct the LSTM model, 58 advantageous in predicting dengue in most cities. Moreover, the model helped to 59 predict the dengue outbreaks in 2018 and used to compare the estimated dengue cases 60 with the RMSE results of other previously published models. A thorough search of 61 the literature shows that this is the first established dengue forecast model using the 62 LSTM method, which is effective in predicting the trend of dengue dynamics. 63 4 64

A Deep Learning Approach for Dengue Fever Prediction in Malaysia Using LSTM with Spatial Attention

“…They observed that this accuracy level resulted from the increase in the number of iterations contributing to the decrease in the root mean squared error (RMSE). Xu et al [16] developed a timely and accurate forecasting model for dengue fever cases in China using LSTM. The researchers compared the performance of LSTM models with other previously published models when predicting dengue fever cases one month into the future.…”

Dengue Cases Prediction Using Machine Learning Approach

“…According to the empirical research, the proposed strategy is a good one, according to the results of the experimental study. Xu et al (2020) developed an efficient and effective dengue forecasting model using LSTM recurrent neural networks (Xu et al, 2019). Only monthly dengue cases and climate conditions were considered in the study.…”

“…There are some symptoms, including a high fever and a skin rash. Bleeding, low blood platelet levels and blood plasma leakage or shock syndrome can occur in a tiny percentage of cases (Xu et al, 2019;Zhang, Su, & Chen, 2021). When a mosquito assaults someone with dengue fever, they've been nibbled by a mosquito carrying the dengue worm.…”

OE-CTST: Outlier-Embedded Cross Temporal Scale Transformer for Weakly-supervised Video Anomaly Detection

“…Thus, such methods fail to detect long anomalies as they do not capture the long-range temporal dependencies. With the recent success of transformers in computer vision which are empowered by multi-head selfattention [9], many popular methods in fully-supervised action detection [17,40] and classification have leveraged temporal transformers for effective global temporal relation encoding. However, weakly-supervised anomaly detection tasks can not get direct benefits from the current temporal transformers, due to (i) conventional positional encoding: unlike fully-supervised settings, where the temporal positions have a one-to-one correspondence with the anomaly instances for superior global temporal relation encoding, the weakly-supervised methods do not have such correspondences due to the unavailability of the instance labels; (ii) naive tokenization scheme: existing methods follow a fixed-scale tokenization scheme regardless of the action duration, as a result these methods can not accumulate local contextual information for long anomalies.…”