Clairvaux

Prediction and visualization of armed conflict events.

Comment 22

Synopsis

This project aims to predict and visualize armed conflict data for African states. Realtime data can be obtained from ACLED (Armed Conflict Location and Event Data Project). Prediction module uses Numenta's Hierarchical Temporal Memory (HTM) model, which is based on a brain-inspired theory of neuroscience.

A co-occurence matrix based on d3.js is used for predictive visualization. This type of matrix is normally used for text-mining. For this project, rows correspond to actual events while columns are predicted events. Darker cells indicate more frequent co-occurrence of actual and predicted events.

Motivation

The primary motivation is to learn about Numenta's HTM framework. A secondary motivation is to explore techniques for categorical prediction and predictive visualization.

Built With

Share this project:

Updates

posted an update —

I modified the inputs to better represent the data file.

  1. GWNO - country code, not encoded and used to reset event sequences
  2. Event Date - day of the event encoded as date
  3. Event type - type of conflict event, encoded as category
  4. Interaction - code to indicate interaction between actors involved in the event, encoded as category

The HTM is learning sequence of events per country. As a side effect, memory usage is reduced slightly and training time is a bit faster.

Log in or sign up for Devpost to join the conversation.