Our AI is the Only Tech in the World that Predicts Future Macro Disruptions


Think labor strikes, massive jumps in commodities, currencies and more.


7 of the Fortune 100 already do.






CASE STUDY: PREDICTING THE NEXT STRIKE




Total # of labor strike events that
SELDN predicted in advance,
from March 2015 to August 2015

*Interact With Our Visuals

Labor Strike Predicted (90% Accuracy) — August 2015 No Labor Strike Predicted (95% Accuracy) — August 2015

SELDN Labor Strikes Prediction: Case Study

The ability to predict labor strikes is essential to the development and productivity of supply chains. In our case study, we use GDELT and the World Bank's WDI dataset to make labor strike predictions for 12 different countries. From March 2015 to August 2015, Seldn was capable of accurately predicting 62 labor strike events in advance. With this knowledge, supply chains would have the capability of taking precautionary measures to mitigate any potential losses that may have resulted from the labor strikes. For example, the West Coast Port Strikes of 2012 possessed an economic cost of nearly $1 Billion USD per day. Being aware of future labor strikes can allow a supply chain to have alternate shipment routes, cover shortages, and reduce or avoid vital part shortages and stock-outs.

SELDN Strike Prediction Record

SELDN Prediction Accuracy

HOW SELDN PREDICTS LABOR STRIKES

Seldn's Labor Strike 1.0 model is built to predict which countries will and will not experience labor strikes in each month, 6 months into the future. Initial predictions for strikes have been conducted using a target dataset of global labor strikes. The Dataset was created using Natural Language Processing of global media coverage (See the GDELT project). Only GDELT and the World Bank's WDI dataset were used in the model. These predictions are limited to twelve countries (See list below), due to a lack of recent data across a broad number of development indicators from other countries in the WDI data set necessary for accurate prediction.

RESULTS

When this model predicts a Strike, it is correct 4 out of 5 times. When it predicts a No-Strike, it is correct 9 out of 10 times. Current predictions have an overall accuracy of 88% with a p-value of 81%. This compares favorably against the benchmark of always making the safe prediction - 85%.

Layer 1 STRIKE PREDICTED, TRUE STRIKE PREDICTED, FALSE NO STRIKE PREDICTED, TRUE NO STRIKE PREDICTED, FALSE NO STRIKE PREDICTED STRIKE PREDICTED

What's the Next Supply Disruption?


FACTORS

Different models have slightly different factors; however, they do share a lot of common indicators, such as population demographics, percentage of imports and exports, and measures of economy [ex. GDP, GNI, etc.].
Our best model indicates that the socioeconomic factors are country specific, only if the country is Russia (i.e., that is the underlying socioeconomic factors around a labor strike is the same unless the country is Russia).
Other interesting correlations, include number of mobile subscriptions and the number of Labor Organizationals making threats in Media. Time is important. Fascinatingly, a lot of the models use time to make their decisions. This probably means that there is a specific year where the socioeconomic factors determining a strike uprising change before and after that tipping year.

METHODS & MODEL

Seldn's automated prediction engine uses a combination of modified classifiers and regressors to predict jumps. Our algorithms run through a total of 12 different models, with a prediction accuracy between 77-89%. Predictions were made monthly up to 5 months out in the future, without any loss in prediction accuracy.

Labor Strike Feature Models vs. Precision and Recall

NEXT STEPS

Seldn is working on incorporating sector, city and industry-specific information to our labor strike predictions.

Our AI Scans the World's Data Constantly.