Machine learning will transform investment management. However, many investment professionals are still developing their understanding of how machine learning works and how you can apply it. With this in mind, what follows is an introduction to machine learning training methods and a flowchart for machine learning decision making with explanatory footnotes that might help determine which approach to make use of based on the top goal.
Machine learning training methods
1. Ensemble learning
No matter how fastidiously chosen, every machine learning algorithm has a certain margin of error and is susceptible to noisy predictions. Ensemble learning addresses these shortcomings by combining predictions from different algorithms and averaging the outcomes. This reduces noise and subsequently results in more accurate and stable predictions than the very best single model. In fact, ensemble learning solutions have won many prestigious machine learning competitions through the years.
Ensemble learning groups together either heterogeneous or homogeneous learners. Heterogeneous learners are several types of algorithms combined with a voting classifier. In contrast, homogeneous learners are mixtures of the identical algorithm based on different training data bootstrap aggregation, or sagging, technique.
2. Reinforcement learning
As virtual reality applications increasingly resemble real-world environments, trial-and-error machine learning approaches may be applied to financial markets. Reinforcement learning algorithms distill insights through interaction with one another and from data generated by the identical algorithm. They also use supervised or unsupervised deep neural networks (DNNs) in deep learning (DL).
Reinforcement learning made headlines when DeepMind’s AlphaGo program defeated the reigning world champion in the traditional game of Go in 2017. The AlphaGo algorithm features an agent that performs actions that maximize rewards over time while respecting the constraints of its environment.
Reinforcement learning with unsupervised learning has neither directly labeled data for every statement nor immediate feedback. Rather, the algorithm must observe its environment, learn by testing latest actions – a few of which might not be immediately optimal – and reapply its previous experiences. Learning happens through trial and error.
Scholars and practitioners apply reinforcement learning in investment strategies: The agent may very well be a virtual trader who follows certain trading rules (actions) in a certain market (environment) to maximise his profits (rewards). However, it continues to be an open query whether reinforcement learning can handle the complexity of monetary markets.
Machine learning decision making flowchart
Footnotes
1. Principal component evaluation (PCA) is an indicator of the complexity of the prediction model and helps reduce the variety of features or dimensions. When the information has plenty of highly correlated data XI features or inputs, then a PCA can perform a base shift on the information in order that only the principal components with the very best explanatory power on the variance of the features are chosen. Numerous N linearly independent and orthogonal vectors – by which N is a natural number or a non-negative integer – known as base. Inputs are features in machine learning, while inputs in linear regression and other traditional statistical methods are called explanatory or independent variables. Also a goal Y (Output) in machine learning is an explained or dependent variable in statistical methods.
2. Natural language processing (NLP) includes, amongst other things, sentiment evaluation of text data. It typically involves multiple supervised and unsupervised learning steps and is commonly considered self-supervised since it has each supervised and unsupervised properties.
3. Simple or multiple linear regression without regularization (penalization) is normally categorized as a standard statistical technique but not as a machine learning method.
4. Lasso Regression or L1 Regularization and Ridge Regression or L2 Regularization are regularization techniques that use penalization to forestall overfitting. Simply put, Lasso is used to scale back the variety of features or feature selection while Ridge maintains the variety of features. Lasso tends to simplify the goal prediction model, while Ridge may be more complex and handle multicollinearity in features. Both regularization techniques may be applied not only to statistical methods, including linear regression, but additionally to machine learning equivalent to deep learning to cope with nonlinear relationships between targets and features.
5. Machine learning applications that use a deep neural network (DNN) are sometimes called deep learning. Target values ​​are continuous numerical data. Deep learning has hyperparameters (e.g. variety of epochs and regularization learning rate) which can be specified and optimized by humans, not by deep learning algorithms.
6. Classification and regression trees (CARTs) and random forests have goal values ​​which can be discrete or categorical data.
7. The variety of clusters K – considered one of the hyperparameters – is an input from a human.
8. Hierarchical clustering is an algorithm that groups similar input data into clusters. The variety of clusters is set by the algorithm, not by direct human input.
9. The K-Nearest Neighbors (KNN) algorithm may also be used for regression. The ANN algorithm requires a set of neighbors (classifications) provided by a human as hyperparameters. The KNN algorithm may also be used for regression, but it surely is omitted for simplicity.
10. Support Vector Machines (SVMs) are sets of supervised learning methods which can be applied to linear classification but additionally use nonlinear classification and regression.
11. Naive Bayes classifiers are probabilistic and apply Bayes’ theorem with strong (naive) independence assumptions between features.
References
If you enjoyed this post, remember to subscribe.
Photo credit: ©Getty Images/Jorg Greuel
