How Our Predictive Models Are Created

To achieve the highest possible success rate in our short-term and long-term predictive models, we utilise a large amount of data from various sources and predictive methods. Our predictive models are based on many global and local models, artificial intelligence, statistical evaluations, and other available data at the given time. Our data is derived from internal sources of various companies, statistical offices, universities, research institutions, and primarily meteorological services.

Most meteorological services around the world prepare their forecasts based on the output of a single global model (usually the American GFS or the European ECMWF) and refine the data locally using methods like WRF. Regardless of the accuracy of the local model settings, the limitation lies in the accuracy of the chosen global model. However, we employ a different and rather unique approach to eliminate as many errors as possible and achieve maximum prediction accuracy. As a result, our predictive models are highly accurate, leading to significant achievements in both the B2B and B2C sectors.

Weather forecast using machine learning. — How we create accurate weather forecasts using our ML models, which then influence our predictive models.

ML Combination of Global Models

First, we utilise our extensive database of past weather and archived runs of numerical models to create a unified global output, which eliminates most biases and errors. Our machine learning algorithms evaluate the past errors of individual models under different meteorological situations and locations. The output provides the most probable scenario for weather development based on a combination of all available models. This output is calculated four times a day.

Custom Local Meteorological Predictive Model and AI Weather Correction Models

The global multimodel then serves as the basis for our regional models, which further increase resolution and consider local specifics and geography. At this stage, the meteorologist on duty can also adjust the input based on their experience.

For locations where measurements from meteorological stations are available, typically in cities and more populated areas, we utilise an additional model layer that uses artificial intelligence (AI) to correct the outputs based on the specific characteristics of the area and meteorological conditions.

All this data plays a crucial role in creating individual applied predictive models. For example, in combination with internal data from individual companies and other sources, we can accurately predict demand, customer purchasing behaviour, optimise inventory creation, better control the production process, save costs, and realise higher profits for any organisation. We look into the past, perceive the present, and can read the future with great precision.

Real-Time Nowcasting

Due to the large variability of certain data, in some situations, it is necessary to adjust even our best models based on the current actual developments. For example, for individual weather-related parameters, we utilise measurements (from stations, radars, etc.) available for the given area and update our outputs according to the latest weather developments. As weather significantly influences mood and consumer behaviour, it is crucial to include it as a significant data source in our comprehensive predictive models. This enables us to offer the most up-to-date predictive model available, hourly and even minutely, for various business areas within the B2B and B2C segments.

Data Update Frequency

Shopping behaviour and demand can change as quickly as the weather. At Forecasts.Cloud, we merge all available data and current states of various variables into our predictive models in real-time and update the forecasts as needed. For most locations, the data is updated every 10 minutes.

Current weather and minute-by-minute forecast are based on available measurements and updated as soon as the information is available, at least once an hour. For certain locations, the data is updated in 10-minute intervals.

We have unique applied predictive models

Our predictive models are derived from multiple data sources. With these data and our expertise in machine learning, programming, meteorology, mathematics, analytics, and business, we create unique applied forecasting models for your business.

We can predict not only demand but also your customers' behavior, cost estimates, optimize advertising campaigns, or even website and store traffic. Examples of our indicators:

Demand indicators
Customer shopping behavior indicators
Occupancy indicators in various industries
Marketing indicators associated with efficiency
Business indicators associated with costs
Website and application traffic indicator
Effectiveness indicator of online advertising campaigns
Engagement indicator in the online environment
And other specific indicators...