Мы используем файлы cookie для быстрой и удобной работы сайта. Выберите, какие файлы cookie вы разрешаете нам использовать. Подробнее в Политике конфиденциальности.
 
We use cookies to ensure our website works quickly and easily. Choose which cookies you allow us to use. Learn more in our Privacy Policy.
 
Compilation of expert assessments of vendors on gender and age prediction from CCTV cameras
AI-based systems
system integrators
7%
5%
the accuracy of gender prediction has increased
by
by
the accuracy of age prediction has increased
Customer
A large IT department of an executive authority, Moscow region
CHALLENGES/FEATURES
Increasing forecast accuracy required fine-tuning of the models, since the vendors' accuracy was already relatively high
To improve the accuracy of the vendor assessment set, it was necessary to apply a model, but the nature and type of the model had to be determined at the implementation stage
Improving the accuracy of gender and age prediction based on responses from 4 vendors in CCTV images
Task
solution
Technical solution
1. Bringing expert system assessments to a common format for ensemble
2. Developed an algorithm for combining the assessment of expert systems for predicting gender and age using video images from CCTV cameras
3. We evaluated the performance of the ensemble system using test data.
Ensembling is a machine learning method that combines multiple models (expert estimates) to improve forecast quality. Each individual model has strengths and weaknesses, and ensemble allows for increased stability of the developed model and improved forecast accuracy.
4. Optimized the use of a limited amount of labeled data. Machine learning algorithms and active learning techniques were applied to improve forecasting accuracy and efficiency.
5. Implemented the developed ensemble system
6. The following approaches were used:
weighted voting
an ensemble method where each model in the ensemble is weighted according to its importance in decision making
algebraic methods
allow you to process data, identify patterns, reduce data dimensionality, build models, and make decisions based on algebraic calculations
neural networks
class of machine learning algorithms
approximation by hyperplanes
Result
Business values
Increased accuracy
Gender prediction accuracy increased by 7%
Age prediction accuracy has increased by 5%
The developed systems estimate gender and age based on multiple expert assessments, which allows for more reliable results.
Optimizing the use of limited data
One of the key features of the project is the ability to use only final estimates obtained from expert systems, which reduces the likelihood of errors and increases the accuracy of estimates.
By clicking the "Submit" button, you expressly consent to the processing of your personal data to the extent and for the purposes defined in the Personal Data Processing Policy.
Development of software
and Big Data solutions
Send a request and our specialists will contact you within 1 hour.
Choose a convenient method of communication
You can attach three files up to 3 MB each. Formats: doc, docx, pdf, ppt, pptx
Сообщение об успешной отправке!