Symilarity’s Location Prediction Software enables any business to undertake location event forecasting and prediction without specialist resource or hardware. You just need historic data that can be used to create a model.
Example: Predicting Spread of H1N1 Virus (Swine Flu)
- Symilarity’s Location Prediction software uses machine learning to assimilate geospatial time series data and make forward predictions.
- Generalised model – not domain specific. It can potentially apply to any spatial event data (e.g. crime, earthquakes, consumer behaviour, risk, epidemiology etc.)
- State of the art approach – performs non-parametric estimation of intensities at the same time as colouring the events
- Clever optimisation – traditional models will not scale to large datasets because computational complexity is a function of the square of the number of events.
- Symilarity uses clever optimisation to reduce the processing time and uses simple geometry as an approximation
- Enables the use of other numeric event characteristics to be included in the model. These co-variates could be anything upon which Euclidean distance is meaningful e.g. the magnitude of the tremor, the monetary value of the purchase, the number of police on patrol etc. All of these things may have a bearing on the two intensity functions
- An arbitrary number of co-variates may be used at linear computational cost
- Takes user data in simple CSV format and outputs to CSV format
- Enables the user to have separate inference data and prediction data
- Enables the user to specify the prediction time horizon
- Provides parameters to enable the user to tune the model
- Enables the user to weight the relevance of other columns
- Choice of approximation algorithms
- Enables to the user to specify the number of simulation runs required as output, with each simulation being equally likely
- Future probabilities can be visualised using heatmaps as individual simulations or assessed as an empirical average across a number of simulations