Browsing by Author "Wessels, Zander"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Interpreting decision boundaries of deep neural networks
    (Stellenbosch : Stellenbosch University, 2019-12) Wessels, Zander; Lamont, M. M. C.; Reid, Stuart; Stellenbosch University. Faculty of Economic and Management Sciences. Dept. of Statistics and Actuarial Science.
    ENGLISH ABSTRACT: As deep learning methods are becoming the front runner among machine learning techniques, the importance of interpreting and understanding these methods grows. Deep neural networks are known for their highly competitive prediction accuracies, but also infamously for their “black box” properties when it comes to their decision making process. Tree-based models on the other end of the spectrum, are highly interpretable models, but lack the predictive power with certain complex datasets. The proposed solution of this thesis is to combine these two methods and obtain the predictive accuracy from the complex learner, but also the explainability from the interpretable learner. The suggested method is a continuation of the work done by the Google Brain Team in their paper Distilling a Neural Network Into a Soft Decision Tree (Frosst and Hinton, 2017). Frosst and Hinton (2017) argue that the reason why it is difficult to understand how a neural network model comes to a particular decision, is due to the learner being reliant on distributed hierarchical representations. If the knowledge gained by the deep learner were to be transferred to a model based on hierarchical decisions instead, interpretability would be much easier. Their proposed solution is to use a “deep neural network to train a soft decision tree that mimics the input-output function discovered by the neural network”. This thesis tries to expand upon this by using generative models (Goodfellow et al., 2016), in particular VAEs (variational autoencoders), to generate additional data from the training data distribution. This synthetic data can then be labelled by the complex learner we wish to approximate. By artificially growing our training set, we can overcome the statistical inefficiencies of decision trees and improve model accuracy.
  • Thumbnail Image
    Item
    A Walk-Forward Multi-Factor Machine Learning Investment Process
    (Stellenbosch : Stellenbosch University, 2024-12) Wessels, Zander; Engelbrecht, Andries P.; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.
    It can be said that traditional asset management modelling lacks true empiricism due to the difficulty of accurate simulation of investment strategies through time. This is often demonstrated by positively biased academic results that do not reflect real-world outcomes. This thesis addresses this problem by developing a walk-forward platform that is inspired by the blackboard-expert architecture and that can simulate investment processes through time, providing reliable repeatability, which is the essence of good science. The proposed platform accounts for common biases, such as survivorship bias and forwardlooking bias, to create a true hypothesis-testing engine. When considering the investment process as it pertains to equities, this thesis argues that the process can be fully closed under the questions of “what should I buy?”, “when should I buy?”, and “how much should I buy?”. The thesis then aims to use the hypothesis engine to build an example investment pipeline that can test and automate a whole investment process. The proposed investment pipeline will answer the questions of “what”, “when”, and “how much”, respectively. Traditional factor models and machine learning models for stock selection will be explored to answer the “what” question. The thesis argues that training models using the proposed engine is the correct way to do so in a non-stationary time series setting. The “how much” question will be addressed using portfolio optimisation, with particular consideration given to particle swarm optimisation. The “when” question will be briefly discussed as a further research idea.