What is an application or an app?

This article defines an application (app) in the AI & Analytics Engine

An app is a space created on the Engine for experimentation and operationalization of Data Science, Analytics, or Machine Learning (ML) tasks. During the app creation process, users are asked to define a problem with their dataset. Currently, the following two problem types are available in the Engine:

1. Predicting the selected column from a dataset (Classification & Regression)

2. Finding groups of similar items by analyzing patterns in data (Clustering)

Within an app, users can train a single model, or multiple models (run algorithms) for the same task, and compare the results, so they can choose the best-suited result for their next step. Depending on the app type (prediction or clustering), different features and functions are available.

Classification and Regression (Prediction)

For Classification and Regression apps, a single train / test split is created from the dataset used to define the problem, common to all models to be trained. This allows users to compare their performance metrics fairly.

Classification and Regression apps contain:

• Feature Sets

• Models

• Deployments

The user can define multiple feature sets, or sets of columns from which the user chooses to define the inputs to their model. They can also deploy trained models as a prediction API and make batch predictions with a tabular data file or make predictions real-time via the API endpoint.

Clustering (finding groups)

Users define a clustering app by choosing the dataset they want to run clustering on, followed by choosing the columns that define similarity for their purpose. Under Clustering apps, the clustering algorithm chosen at the time of creating the app are run on the entire data immediately, without any train/test split. Clustering apps produce a detailed result page for each algorithm run, containing the following insights:

• Overview: Number of clustered items, number of clusters, size of each cluster

• Analysis of the important dimensions that differentiate clusters

• A detailed description of the clusters in terms of the columns chosen

In addition, users can also download the clustered data or export it as a dataset to the same project.

Templates

Templates allow users to go through an entire workflow of predefined business problems.

Using a template is an efficient and quick way to utilise the power of AI to solve specific problems in a guided manner.

As an example, consider the customer-churn template. The goal of this template is to create a model that can predict, on an periodic basis, the likelihood of churn (the risk) for the active customers of a company, based on their transactional data.

The template guides the Engine’s user to provide details about their business problem and builds all the necessary stages of the ML workflow for the user. Specifically, it guides the user to provide connections to their datasets, the specification of the churn definition, and domain-knowledge inputs about contributing factors. It then automatically engineers predictive features from the data and trains the best models for the churn prediction. The user can consume the results in many ways, such as analysing the root causes of churn or deploy the trained model/s and schedule periodical predictions , thereby getting informed in an ongoing basis about the risk of churn of existing clients, and take preventive actions if needed.