Model Types

Model Categories

Regression, time series forecasting, and classification are three common machine learning models that Akkio can generate. They are generic categories that each have specific models that are used to achieve their goals. At Akkio, we pick from several models to determine which is best for your Regression Time Series or Classification needs.

Model Types

Many models can be used for multiple tasks. For example, neural Networks and Random Forests can be used for Classification and regression. Below is a list of some of the models Akkio uses to give a better insight into what the platform is doing when it picks a model for your data. Akkio also has a proprietary baseline model for both forecast and prediction modeling.

Classification

Machine learning uses classification to predict a categorical label or class. A classification model learns the relationship between input variables and a discrete output variable, such as classifying an email as spam or not, predicting whether a customer is likely to churn. Classification models are trained using labeled data and can be used for tasks such as image recognition, sentiment analysis, or fraud detection. Common algorithms used for classification include logistic regression, decision trees, random forests, support vector machines (SVM), and deep neural networks.

Decision Trees: Learn a hierarchical structure of if-then rules based on input variables. The model is trained using labeled data and can handle categorical and continuous variables. Decision Trees are intuitive and easy to interpret, making them popular for many applications.
- XGBoost: Extreme Gradient Boosting introduces decision tree ensembles. By combining decision trees and gradient boosting, XGBoost creates a series of models that correct errors made by previous models.
Random Forests: An ensemble of decision trees. The model combines multiple decision trees to improve performance and reduce overfitting. Random Forests can handle categorical and continuous variables and are widely used for image classification or stock price prediction tasks.
Neural Networks: Inspired by the structure and function of the human brain. It consists of many interconnected processing nodes or neurons that work together to learn patterns in the input data and make predictions or classifications.
- Dense Neural Network: Each neuron in a layer is connected to every neuron in the previous layer. This means that the weights and biases of each neuron can affect the output of every other neuron in the network. Dense neural networks are commonly used for image and speech recognition, natural language processing, and other complex tasks.
- Sparse Neural Network: Only a subset of the neurons are connected. This means that the weights and biases of each neuron only affect a small subset of the other neurons in the network. Sparse neural networks can be more efficient than dense neural networks for specific tasks, such as recommendation systems or graph-based data.
- Multilayer neural network: Has more than one hidden layer. Each additional hidden layer allows the web to learn more complex patterns in the input data and can improve the accuracy of the predictions or classifications. However, adding more layers can also make the network more difficult to train and more prone to overfitting.
- Deep Neural Network with Attention: Selectively focuses on essential parts of input data during training and inference. It consists of multiple layers of neurons that gradually extract higher-level features. The attention mechanism dynamically weighs the importance of different parts of the input data at each decoding step, improving the model's accuracy and efficiency.

Regression

Use to predict a continuous numerical value. A regression model learns the relationship between input variables and a continuous output variable, such as predicting a person's salary based on their years of experience, age, education level, etc. Regression models are trained using labeled data, where the correct outputs are known and can be used for tasks such as predicting housing prices, stock prices, or the likelihood of a customer buying a product.

Many of the same models listed in Classification are also used for Regression.

Linear Regression: Learns a linear relationship between input variables and a continuous output variable. It is a simple and widely used model that can be trained using labeled data to predict values such as a house's price or a person's weight based on height.

Time Series Forecasting

Time series forecasting predicts future values of a variable over time, given historical data. This is often used in weather forecasting, stock market analysis, and energy demand forecasting. Models are trained using historical data and aim to identify patterns or trends in the data to make accurate predictions about future values. Common algorithms used for time series forecasting include ARIMA, SARIMA, and LSTM neural networks.

Exponential Smoothing: Calculates a weighted average of past observations, with the weights decaying exponentially over time. This technique is often used to smooth out noisy time series data and make short-term predictions. There are different variations of exponential smoothing, including simple exponential smoothing, double exponential smoothing, and triple exponential smoothing.
Auto-Regressive Ensemble: Trains multiple models on lagged versions of the same time series data. Each model predicts a future time step based on its own set of lagged inputs, and the final prediction is obtained by combining the individual predictions of all the models. This approach can improve the accuracy of time series forecasting by capturing complex dependencies between different time steps.
Prophet: Combines additive and multiplicative time series models with seasonality. The model is designed to handle a range of time series patterns, including trend changes, seasonal effects, and holiday effects. Prophet also allows for the inclusion of external factors such as holidays, weather, or economic indicators to improve the accuracy of the forecasts. The model is easy to use and can be trained quickly on large datasets.

PreviousInsights Report - Forecasting NextDeploy

Last updated 1 year ago

Was this helpful?

Classification

Decision Trees: Learn a hierarchical structure of if-then rules based on input variables. The model is trained using labeled data and can handle categorical and continuous variables. Decision Trees are intuitive and easy to interpret, making them popular for many applications.

XGBoost: Extreme Gradient Boosting introduces decision tree ensembles. By combining decision trees and gradient boosting, XGBoost creates a series of models that correct errors made by previous models.

Random Forests: An ensemble of decision trees. The model combines multiple decision trees to improve performance and reduce overfitting. Random Forests can handle categorical and continuous variables and are widely used for image classification or stock price prediction tasks.

Neural Networks: Inspired by the structure and function of the human brain. It consists of many interconnected processing nodes or neurons that work together to learn patterns in the input data and make predictions or classifications.

Dense Neural Network: Each neuron in a layer is connected to every neuron in the previous layer. This means that the weights and biases of each neuron can affect the output of every other neuron in the network. Dense neural networks are commonly used for image and speech recognition, natural language processing, and other complex tasks.
Sparse Neural Network: Only a subset of the neurons are connected. This means that the weights and biases of each neuron only affect a small subset of the other neurons in the network. Sparse neural networks can be more efficient than dense neural networks for specific tasks, such as recommendation systems or graph-based data.
Multilayer neural network: Has more than one hidden layer. Each additional hidden layer allows the web to learn more complex patterns in the input data and can improve the accuracy of the predictions or classifications. However, adding more layers can also make the network more difficult to train and more prone to overfitting.
Deep Neural Network with Attention: Selectively focuses on essential parts of input data during training and inference. It consists of multiple layers of neurons that gradually extract higher-level features. The attention mechanism dynamically weighs the importance of different parts of the input data at each decoding step, improving the model's accuracy and efficiency.

Regression

Many of the same models listed in Classification are also used for Regression.

Linear Regression: Learns a linear relationship between input variables and a continuous output variable. It is a simple and widely used model that can be trained using labeled data to predict values such as a house's price or a person's weight based on height.

Time Series Forecasting

Exponential Smoothing: Calculates a weighted average of past observations, with the weights decaying exponentially over time. This technique is often used to smooth out noisy time series data and make short-term predictions. There are different variations of exponential smoothing, including simple exponential smoothing, double exponential smoothing, and triple exponential smoothing.

Auto-Regressive Ensemble: Trains multiple models on lagged versions of the same time series data. Each model predicts a future time step based on its own set of lagged inputs, and the final prediction is obtained by combining the individual predictions of all the models. This approach can improve the accuracy of time series forecasting by capturing complex dependencies between different time steps.

Prophet: Combines additive and multiplicative time series models with seasonality. The model is designed to handle a range of time series patterns, including trend changes, seasonal effects, and holiday effects. Prophet also allows for the inclusion of external factors such as holidays, weather, or economic indicators to improve the accuracy of the forecasts. The model is easy to use and can be trained quickly on large datasets.