Israeli Students Predict Parliament Voting Patterns

This post is also available in: עברית (Hebrew)

Artificial intelligence students at the Israeli Technion predict voting patterns for Knesset members.

New algorithms can predict the way Knesset members will vote on various laws being passed. Students at the Haifa Technion managed to predict voting patterns for Knesset members with 80% accuracy. They did it by using a computerized system that analyzes parameters from previous voting sessions and predicts how Knesset members – as individuals and as part of their Knesset faction.

The system was designed by students Igal Kreitchman and Nadir Izrael, BSc students at the computer sciences faculty, as part of their artificial intelligence classes under the supervision of Prof. Shaul Markovich.

The students explained that their idea was to build a model that would allow Knesset members getting ready to present a new law proposal to predict how each of their parliament colleagues will respond, the goal being improving the chances of the law getting approved. The students were looking for a challenging issue with a large database of existing answers for a particular question. They found the database on the “open Knesset” internet website, a platform that presents the public with data on Knesset activity, including votes of members and their factions.

iHLS – Israel Homeland Security

Their model uses a certain percentage of past examples – law proposals that were approved by earlier parliaments – studies them and conducts a prediction simulation based on the results. There were several difficulties, such as the difference between the original draft brought forward and the final draft voted on after the third Knesset reading. The difference, it turns out, is sometimes significant enough to cause the Knesset members who presented the original draft to avoid voting on their own proposal. There are also coalition pressures that can change voting preferences, personal considerations of Knesset members and lawmakers changing parties.

Another difficulty – the tendency of lawmakers to be absent during voting sessions. The students discovered that, on average, 19.95 Knesset members are present during voting sessions, less than a sixth of the total number of 120. Another interesting find: Israeli Knesset members have a very strong loyalty to their faction. Very few have voted against their faction, for example. The system successfully predicted votes by individual lawmakers with 80% accuracy, and faction voting patterns with 67% accuracy.

Prof. Markovich said that the early results of the project show promise. The system’s accuracy, according to him, could be improved in the future. “The students located a good database, analyzed it and studied the data using various algorithms.”