Octillion's mission is to transform businesses into intelligent enterprises. Businesses use information infrastructure products which were primarily designed to integrate various data silos and provide an unified view to the corporate management. However, these products lack ability to do sophisticated analysis. IT vendors are now busy retrofitting their products with advanced analytics to add value to decision support system. Octillion team is highly qualified to develop analytics products to integrate with existing general purpose products.

Octillion uses a range of techniques, such as linear regression, neural networks, support vector machines, time series techniques etc for solving these problems. There is no single technique that is good for all problems, as each of the methods has their unique advantages in comparison with others. Following is a brief description of some of the techniques used in Octillion.

Linear Regression: This is a statistical approach to model the dependency of multiple variables to predict a single output variable. The underlying model is defined as a linear relationship between the inputs and output that fits a straight line to a cluster of points. The model is fairly simple, hence cannot fit complex relations with high accuracy. But it is the basis for other advanced techniques and can perform really well in sparse data sets.

Moving Average: In this approach a subset window of fixed size is moved on the complete dataset. The average evaluated in the each of the window bin is used in building the models. The objective of using a moving window is to smooth the data and remove noise. When the moving average is applied to time series data it is referred to as autoregressive moving average (ARMA). There are different variations of ARMA models based on the lag definitions and input variables used.

Support Vector Machine: SVM is a combination of multiple linear regressions, where the space is divided into hyper planes using multiple linear cuts. This can also be further extended to use kernel functions and provide nonlinear fit capabilities.

Artificial Neural Network: ANN is based on the mathematical models of biological neural networks and are capable of mapping complex nonlinear functions. The mean squared error is used as the learning function to adapt the weights of the neurons and provides a high accuracy in prediction on training data.