Algorithm Technologies

Machine learning algorithms are used to find natural patterns in datasets to support better decision making and prediction.

• Type prediction

  The email system uses machine learning classification algorithms to classify received emails into genuine emails and spams to improve user experience.

  

• Data prediction

  The meteorological bureau uses machine learning regression algorithms to analyze data including recent temperature, humidity, and wind direction to predict the temperature.

  

• Pattern mining

  Retailers use pattern mining algorithms to analyze customers' purchase behavior, finding what products (such as milk and bread) are often bought at the same time.

  

When Do I Need to Use a Machine Learning Technology?

1. Complex tasks or problems involving a large amount of data or multiple factors
2. Scenarios that have no specific handling formulas or service rules.

Specifically, for fraud detection among transaction records, the task mode keeps changing and there are many factors that affect the detection result, which makes it difficult to cover all cases with specific rules. In automatic transaction and shopping trend prediction scenarios, the data mode and service rules keep changing and maintenance is labor-consuming. In text classification and speech recognition scenarios, rules and patterns are too complex to be described using specific rules.

During client-to-end data analysis, ISVs preprocess collected data, for example, selecting data sources, labeling and structuring data, and verifying data validity. The algorithm library implements algorithm model computing. For example, in terms of the supervised learning algorithm, feature engineering is performed on data, and then algorithm model training is performed based on the input data. Then, the algorithm model is output for customers or ISVs to do inference, and the result is visualized.