Algorithms Employed

Our machine learning platform has a growing suite of about 30 core algorithms with a total of over 300 permutations. Given your data, our suite of algorithms will tune and select a model that best fits your data and the problem you’re trying to solve.

Forecasting a future quantity

Solves the question, how much of x can I expect in the future?
  1. Algorithms support hourly, daily, weekly, monthly, and annual seasonalities
  2. All support Anomaly Smoothing and Model Ensembling
    • Various combinations of
    • AutoRegressive component with p parameters
    • Differencing component with d parameters
    • Moving Average component with q parameters
    • with external regressors
  • Exponential Smoothing
    • Simple
    • Double
    • Triple
    • with Box-Cox Transformation
  • Autoregressive Neural Network
    • with or without external regressors
  • Multiple Linear Regression
    • with or without external regressors
  • Spline
  • Seasonal and Trend Decomposition using Loess
    • with ARIMA
    • with or without external regressors
    • with Exponential Smoothing
  • Bayesian Time Series Regression
    • with or without external regressors
  • Additive Model
  • Home-grown Nexosis Algorithms

Predicting a variable

Solves the question, what can I expect x to be?
  • Least Squares
    • Linear
    • Polynomial
  • Elastic Net
  • Lasso
  • Ridge
  • Support Vector Regression
    • Linear Kernel
    • Polynomial Kernel
    • Radial Basis Function kernel
    • Sigmoid Kernel
  • Multi-Layer Perceptron (Neural Network)
    • with 1, 2, or 3 hidden layers
    • Rectified Linear Unit Function
    • Hyperbolic Tan Function
    • Sigmoid Function
  • Random Forest
  • K-Nearest Neighbor
  • Logistic Regression
  • Naive Bayes
  • XGBoost