Reduce Emergency Response time using Machine learning Technique

In reaction to any unusual circumstances, an ordinary man looks for the government’s response system to play its pivotal role in providing a timely response by any respective field force. It is the responsibility of emergency response forces to ensure a rapid and appropriate reaction to any undesired situation. According to statistics obtained from emergency centers in Pakistan and different countries, it is brought into consideration that every day, around 20–25 thousand emergency calls are received, and on weekend’s ratio increases by 20%. Extensively it is classified by different studies, facts, and feedback that 80% of the calls are hoax calls. Currently, the system is entirely dependent on a human resource where a human operator is required to listen to the person in the problem and decide which response force needs to be informed. This naturally brings a delay in response time. The system that is being developed first analyses voice-to-text conversions from a semantic perspective applies data parsing techniques to the text that has been extracted, and then creates tokens based on pre-learned keywords. The system is trained on a data set that was obtained from a well-known emergency command center. Keeping in mind the confidentiality of the data set, only the emergency-related descriptions were used to train our model. For each unique case, a keyword was listed in the database as a token, allowing the system to make decisions independently for each pattern it had learned. We have used SVM and logistic regression algorithms in a python framework for training and testing of research work. SVC will extract features from the data set then a pipelining algorithm will clean the data set and remove the unwanted data. It is a kind of pre-processing that is being done to achieve the best and the better accuracy measures. Then these text tokens will be converted into vectors, and a vectorizer algorithm needs vectors/numbers for input and processing purposes. Then a function will take training data and trained labels in a classifier to predict the data. These prediction values and trained labels will be compared to achieve the best accuracy. Similarly, in the testing, the classification report is generated without the help of the fit function because training is allocated with the fit function, as it is not needed in the testing phase.