Large enterprises harness the achievements of Artificial Intelligence and Machine Learning models to manage large volumes of input data that require unification and analysis of various factors. Our ML models present high accuracy, on condition that an amount of historical data is enough.
AM-BITS has proven expertise in building ML models for various business tasks, such as:
Challenge:
Reduction in operating costs for provisioning cash machines.
Solution:
The actual daily data of ATM cash withdrawal was used for further data analysis. We utilized Gradient Boosting Regressor to build an ML-model .
The solution comprised three steps. The first stage:
The second stage focused on:
The third stage was the implemention of:
Result:
Implementation of automated cash demand forecasting with an error range of 0.01% to 3.5%. Great reduction in operating costs by lowering the amount of allocated funds up to 30%, cashback up to 40%, out-of-cash downtime up to 0.2%.
Challenge:
Obtaining a short-term forecast of energy consumption for procurement planning on the Ukrainian Energy Exchange.
Solution:
While analyzing Machine Learning algorithms and methods for short-term and time-bound forecasting of events, we opted for using Recurrent neural networks (RNN). RNN made possible to process a series of events in time or sequential spatial chains.
To conduct the research and set up the test RNNs, we leveraged the open-access hourly data stats on electricity consumption by New York City alongside temperature fluctuations during this period. As a result of our analysis, we managed to get an accurate forecasting for two days’ consumption rates upfront by using a 3-month range of historical data.
Result:
Utilization of Recurrent neural networks (RNN) to build an effective ML-model. Customers can get short-term forecasts of energy consumption based on the analysis of historical data, with an accuracy of 96.4-99.5%.
Challenge:
The verification of call center customers based on voice data as a part of multi-factor authentication for security improvement.
Solution:
The use of deep neural architecture for comparing voice prints from a database, with verification accuracy of over 90%.