Time series classification
- Azure ML
- Exploratory analysis
Interactive system for multivariate time series data classification and anomaly detection on Azure ML platform
Our developers have designed and developed a system for Multivariate time series data classification and anomaly detection in the data measured over a long time period, as parameters corresponding to correct and incorrect system behavior. The user interface based on web services was realized using Azure ML as well.
The interface allows the users, who have no machine learning knowledge, to upload a new data, retrain the system and get the results of prediction and analysis in a user-friendly form.
The main concern of this task was a big load of data and parameters with a low number of labeled data for each of the classes, for one certain measurement period.
Therefore, the work was completed in a few steps:
- statistical and exploratory analysis of the available data;
- feature extraction for further classification based on the statistical parameters and association rules as a description;
- SVM classifier training using techniques for unbalanced data;
- definition of the necessity of additional data from the user, as well as definition of the potentially erroneous outcome when there is not enough data for decision making based on the anomaly detection methods and evaluation of input data uniqueness in comparison with the training set;
- development of the web services for the experiments described, customization of the system retraining when new data is received via Azure data factory.
Model for restoring blurred and pixelated faces in a photo.
Workout helper app
Mobile app for the estimation of proper body positions during the workout.
Car rental price simulation and prediction
The goal of the project is to train models for car rental price prediction in Japan based on the prices and demand for car rental from some Japanese car rental companies and the history of weather data in Japan.