Effective Flood prediction model based on Twitter Text and Image analysis using BMLP and SDAE-HHNN

Abstract

In recent years, social media platforms such as Twitter have garnered a lot of interest as a new source of text data for quick flood awareness and effective prediction. Hence various types of research were made on flood prediction using Twitter data but it only focuses on classifying the text data as relevant or irrelevant, thereby loss of semantic information from longer phrases while extracting important information from Twitter text data and resulting in low accuracy of text classification. Hence, a novel BMLP and SDAE-HHNN has been proposed. This approach comprises BMLP and SDAE-HHNN techniques has been developed for effective flood prediction based on Twitter text data and image analysis. To classify the text data into two/six different classes, BERT is used to preprocess the text data from Twitter. To achieve high levels of precision, the Rule-Based Matching technique extracts specific place entities from the Named Entity Recognition. To predict the high probability location affected by flood from the place entity, bi-directional MLP (BMLP) is used which is made up of a finite number of sequential layers in its most basic form. Then images are extracted from this particular location and these images are processed to predict flood level but existing techniques cannot provide sufficient information to map the flood area and object detection due to real field data collection. Hence, a novel SDAE with HHNN has been developed in which SDAE removes noise from the specified extracted location and HHNN is used to classify the image into flood or non-flood. Then plot this sufficient predicted information related to flood level in the google map. The proposed model is implemented in the Python platform and the result obtained shows that the proposed has a maximum recall of 96%, maximum precision of 95%, accuracy of 97%, and an F1 score of 96%.