Paper: arXiv 2412.16160
Authors: Adamantios Ntakaris, Gbenga Ibikunle
Abstract
This study presents an autonomous experimental machine learning protocol for high-frequency trading (HFT) stock price forecasting that involves a dual competitive feature importance mechanism and clustering via shallow neural network topology for fast training. By incorporating the k-means algorithm into the radial basis function neural network (RBFNN), the proposed method addresses the challenges of manual clustering and the reliance on potentially uninformative features. More specifically, our approach involves a dual competitive mechanism for feature importance, combining the mean-decrease impurity (MDI) method and a gradient descent (GD) based feature importance mechanism. This approach, tested on HFT Level 1 order book data for 20 S&P 500 stocks, enhances the forecasting ability of the RBFNN regressor. Our findings suggest that an autonomous approach to feature selection and clustering is crucial, as each stock requires a different input feature space. Overall, by automating the feature selection and clustering processes, we remove the need for manual topological grid search and provide a more efficient way to predict LOB’s mid-price.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 7.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced ML concepts (neural network topology, gradient descent, statistical measures) but lacks deep theoretical derivations, while its empirical rigor is supported by testing on real HFT data from 20 S&P 500 stocks with clear performance metrics.
Research Flowchart
flowchart TD A["Research Goal:<br>Automate HFT Stock<br>Forecasting"] --> B["Input: HFT Level 1<br>Order Book Data<br>(20 S&P 500 Stocks)"] B --> C["Methodology:<br>Dual Competitive<br>Feature Importance"] C --> C1["(1) Mean Decrease<br>Impurity MDI"] C --> C2["(2) Gradient Descent<br>Based Importance"] C1 & C2 --> D["K-Means Clustering<br>Integrated into RBFNN"] D --> E["Computational Process:<br>Autonomous Feature<br>Selection & Topology"] E --> F["Key Outcome:<br>Improved Mid-Price<br>Forecasting Accuracy"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes an autonomous experimental machine learning protocol for HFT stock price forecasting that uses dual feature importance (MDI and GD) and k‑means clustering integrated with a radial basis function neural network.
Source sentence
This study presents an autonomous experimental machine learning protocol for high-frequency trading (HFT) stock price forecasting that involves a dual competitive feature importance mechanism and clustering via shallow neural network topology for fast training.
Assumes 4
- The k‑means algorithm assumes isotropic clusters with constant variance. model
Source sentence
we made the assumption of isotropic clusters within the k-means algorithm, which presumes that every cluster exhibits constant variance.
- The forecasting target is the Limit Order Book’s mid‑price. data
Source sentence
the prediction of the Limit Order Book’s (LOB’s) mid-price.
- Features are limited to those engineered from the existing best LOB price level. data
Source sentence
our method utilizes certain specific features that can be easily engineered from the existing best LOB price level.
- The experimental protocol is tailored specifically to the mid‑price prediction task (narrow AI approach). other
Source sentence
the study can be considered a narrow AI approach, which means that the experimental protocol developed is specifically tailored to this particular objective - the prediction of the Limit Order Book’s (LOB’s) mid-price.
Data
- HFT Level 1 order book data for 20 S&P 500 stocks order-book unknown
Source sentence
Every table contains 20 stocks over the period of three months (i.e., from September to November 2022).
Costs
Not applicable
Finds 4
- The MDI method under the Simple feature set achieved the lowest relative RMSE for 36 out of 60 cases across three months per stock. performance
Source sentence
As outlined in Table VII our findings indicate that the MDI method, under the Simple feature set, achieved the lowest RRMSE for 36 out of 60 cases (i.e., across three months per stock).
- Performance varies per stock; regime changes from MDI to GD occur on average every 10 trading events. statistical
Source sentence
on average, there was a regime change from MDI to GD every 10 trading events.
- GOOGL under the GD approach with the Extended feature set achieved the lowest RMSE among all experimental scenarios. predictive-accuracy
Source sentence
For instance, GOOGL under the GD approach and based on the Extended feature set achieved the lowest RMSE score compared to the rest experimental scenarios.
- Stocks exhibit alternating numbers of clusters (two or three) and rapid changes in feature‑importance method performance at nanosecond granularity. statistical
Source sentence
All the stocks exhibited a constant alternation between two and three number of clusters within MDI and a rapid rate of change between the two feature importance methods - on average, there was a regime change from MDI to GD every 10 trading events.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.