The Aim of this project is to try to be able to Predict the vulnerability of various Indian states(with plans of expanding it to other countries) towards the chances of facing a terrorist attack(or high level criminal activity) during a specific year, So as to give us an idea and be better prepared to avoid any disastrous situation. This has been done by Training a Model using the data from "The Global Terrorism Database" by START.inc.
Important points for modeling process:-
- The training data is got from- https://www.start.umd.edu/gtd/
- from the 130 present features, 30 are selected using PCA and blank and invalid cells are treated.
- The categorical values are Label encoded using "data_preprocessing.ipynb" file.
- The model which is trained in "training.ipynb" is a 4 hiddeen layers neural network with ADAM's Optimization.
- The target variable named "Vulnerability" is created with analysis of training data and has 5 possible categorical values as:- -> 0 = 0 or 1 possible attacks -> 1 = 2 to 4 possible attacks -> 2 = 5 to 10 possible attacks -> 3 = 11 to 15 possible attacks -> 4 = more than 15 possible attacks
- The final model is run using the python program "Trained_Model.py".
- The resultant predictions has been plotted using the Folium Map library.
Important Instructions To Run "Trained_Model.py":-
- First Install the Folium Map library using the below code pip install git+https://github.com/python-visualization/folium
- Then run the code using the command- python Trained_Model.py
Note:- Only 20% of the total available dataset of 1.8 lakh entries has been able to be used for training till now due to high analysis requirement. An updated version with all the data used and higher accuracy will be uploaded soon.
