Dymitr Ruta

Quantitative approach to collaborative learning: performance prediction, individual assessment, and group composition

The benefits of collaborative learning, although widely reported, lack the quantitative rigor and detailed insight into the dynamics of interactions within the group, while individual contributions and their impacts on group members and their collaborative work remain hidden behind joint group assessment. To bridge this gap we intend to address three important aspects of collaborative learning focused on quantitative evaluation and prediction of group performance. First, we use machine learning techniques to predict group performance based on the data of member interactions and thereby identify whether, and to what extent, the group’s performance is driven by specific patterns of learning and interaction. Specifically, we explore the application of Extreme Learning Machine and Classification and Regression Trees to assess the predictability of group academic performance from live interaction data. Second, we propose a comparative model to unscramble individual student performances within the group. These performances are then used further in a generative mixture model of group grading as an explicit combination of isolated individual student grade expectations and compared against the actual group performances to define what we coined as collaboration synergy - directly measuring the improvements of collaborative learning. Finally the impact of group composition of gender and skills on learning performance and collaboration synergy is evaluated. The analysis indicates a high level of predictability of group performance based solely on the style and mechanics of collaboration and quantitatively supports the claim that heterogeneous groups with the diversity of skills and genders benefit more from collaborative learning than homogeneous groups.

Robust method of sparse feature selection for multi-label classification with Naive Bayes

The explosive growth of big data poses a processing challenge for predictive systems in terms of both data size and its dimensionality. Generating features from text often leads to many thousands of sparse features rarely taking non-zero values. In this work we propose a very fast and robust feature selection method that is optimised with the Naive Bayes classifier. The method takes advantage of the sparse feature representation and uses diversified backward-forward greedy search to arrive with the highly competitive solution at the minimum processing time. It promotes the paradigm of shifting the complexity of predictive systems away from the model algorithm, but towards careful data preprocessing and filtering that allows to accomplish predictive big data tasks on a single processor despite billions of data examples nominally exposed for processing. This method was applied to the AAIA Data Mining Competition 2014 concerned with predicting human injuries as a result of fire incidents based on nearly 12000 risk factors extracted from thousands of fire incident reports and scored the second place with the predictive accuracy of 96%.

Fast summarization and anonymization of multivariate big time series

Sequential, predominantly temporal nature of the vast amounts of big data released every day from many different sources could potentially be linked, aligned along the time and deliver new evidence for the next generation predictive systems or knowledge discovery engines. However, big data owners are reluctant to share their data due to legally binding privacy and identity protection concerns, thereby posing a major hurdle preventing shared exploitation of big data on a massive scale. Data anonymization is expected to solve this problem, yet the current approaches are limited predominantly to univariate time series generalized by aggregation or clustering to eliminate identifiable uniqueness of individual data points or patterns. For multivariate time series, uniqueness among of the combination of values or patterns across multiple dimensions is much harder to eliminate due the to exponentially growing number of unique configurations of point values across multiple dimensions. Our method implements linearly scalable asynchronous summarization of multivariate time series independently at every dimension. As a result the series retain only a small subset of defining points at different times along multiple dimensions effectively breaking up the multivariate time series into a collection of summarized univariate time series that are perturbed from the original series in terms of actual points and pattern shapes. Current implementation of the anonymizing summarization involves shape preserving greedy elimination and aggregation that supports parallel cluster processing for big data implementation.

Self-Organized Predictor of Methane Concentration Warnings in Coal Mines

Coal mining operation continuously balances the trade-off between the mining productivity and the risk of hazards like methane explosion. Dangerous methane concentration is normally a result of increased cutter loader workload and leads to a costly operation shutdown until the increased concentrations abate.

Autonomous deployment of mobile sensors network in an unknown indoor environment with obstacles

We developed a Voronoi-based algorithm, called Bio-Inspired Self Organizing Network (BISON), designed to provide a successful deployment of wireless sensor network (WSN) following fast, cost-efficient and self-organizing process, autonomously adapting to the unknown topology of the target environment, and avoiding obstacles discovered in real-time. To limit the power consumed during the deployment, BISON restricts each node to use only locally sensed information to adapt to live-discovered topology while avoiding obstacles and connecting with neighboring nodes. The algorithm is evaluated with respect to several metrics, and simulation results showed faster convergence to a fully connected network with lower deployment costs compared to similar algorithms reported in the literature.

Inverted ant colony optimization for search and rescue in an unknown maze-like indoor environment

We demonstrate the applicability of inverted Ant Colony Optimization (iACO) for target search in a complex unknown indoor environment simulated by a maze. The colony of autonomous ants lay repellent pheromones to speed up exploration of the unknown maze instead of reinforcing presence in already visited areas. The role of a target-collocated beacon signal within the maze is evaluated in terms of its utility to guide the search. Variants of iACO were developed, with beacon initialization (iACO-B), and with increased sensing ranges (iACO-R with a 2-step far-sightedness) to quantify the most effective one. The presented models can be implemented with self-organizing wireless sensor networks carried by autonomous drones or vehicles and can offer life-saving services of localizing victims of natural disasters or during major infrastructure failures.

Search in a Maze-Like Environment with Ant Algorithms: Complexity, Size and Energy Study

We demonstrate the applicability of inverted Ant Algorithms (iAA) for target search in a complex unknown indoor environment with obstructed topology, simulated by a maze. The colony of autonomous ants lay repellent pheromones according to the novel local interaction policy designed to speed up exploration of the unknown maze instead of reinforcing presence in already visited areas. The role of a target-collocated beacon emitting a rescue signal within the maze is evaluated in terms of its utility to guide the search. Different models of iAA were developed, with beacon initialization (iAA-B), and with increased sensing ranges (iAA-R with a 2-step far-sightedness) to quantify the most effective one. Initial results with mazes of various sizes and complexity demonstrate our models are capable of localizing the target faster and more efficiently than other open searches reported in the literature, including those that utilized both AA and local path planning. The presented models can be implemented with self-organizing wireless sensor networks carried by autonomous drones or vehicles and can offer life-saving services of localizing victims of natural disasters or during major infrastructure failures.

Algorithmic Daily Trading Based on Experts’ Recommendations

Trading financial products evolved from manual transactions, carried out on investors’ behalf by well informed market experts to automated software machines trading with millisecond latencies on continuous data feeds at computerised market exchanges. While high-frequency trading is dominated by the algorithmic robots, mid-frequency spectrum, around daily trading, seems left open for deep human intuition and complex knowledge acquired for years to make optimal trading decisions. Banks, brokerage houses and independent experts use these insights to make daily trading recommendations for individual and business customers. How good and reliable are they? This work explores the value of such expert recommendations for algorithmic trading utilising various state of the art machine learning models in the context of ISMIS 2017 Data Mining Competition. We point at highly unstable nature of market sentiments and generally poor individual expert performances that limit the utility of their recommendations for successful trading. However, upon a thorough investigation of different competitive classification models applied to sparse features derived from experts’ recommendations, we identified several successful trading strategies that showed top performance in ISMIS 2017 Competition and retrospectively analysed how to prevent such models from over-fitting.

Using Recommendations for Trade Returns Prediction with Machine Learning

Automatically predicting stock market behavior using machine learning and/or data mining technologies is quite a challenging and complex task due to its dynamic nature and intrinsic volatility across global financial markets. Forecasting stock behavior solely based on historical prices may not perform well due to continuous, dynamic and in general unpredictable influence of various factors, e.g. economic status, political stability, voiced leaders’ opinions, emergency events, etc., which are often not reflected in historic data. It is, therefore, useful to look at other data sources for predicting direction of market movement. The objective of ISMIS 2017 Data Mining Competition was to verify whether experts’ recommendations can be used as a reliable basis for making informed decisions regarding investments in stock markets. The task was to predict a class of a return from an investment in different assets over the next three months, using only opinions given by financial experts. To address it, the trading prediction is formulated as a 3-class classification problem solved within supervised machine learning domain. Specifically, a hybrid classification system has been developed by combining traditional probabilistic Bayesian learning and Extreme Learning Machine (ELM) based on Feed-forward Neural Networks (NN). Assuming feature space narrowed down to just the latest experts recommendations probabilistic and ELM classifiers are trained and their outputs fed to train another baseline ELM classifier. The outputs from baseline classifiers are combined by voting at the decision level to generate final decision class. The presented hybrid model achieved the prediction score of 0.4172 yielding \(8^{th}\) place out of 159 teams competing in the ISMIS’ 2017 competition.

An ensemble model with hierarchical decomposition and aggregation for highly scalable and robust classification

This paper introduces an ensemble model that solves the binary classification problem by incorporating the basic Logistic Regression with the two recent advanced paradigms: extreme gradient boosted decision trees (xgboost) and deep learning. To obtain the best result when integrating sub-models, we introduce a solution to split and select sets of features for the sub-model training. In addition to the ensemble model, we propose a flexible robust and highly scalable new scheme for building a composite classifier that tries to simultaneously implement multiple layers of model decomposition and outputs aggregation to maximally reduce both bias and variance (spread) components of classification errors. We demonstrate the power of our ensemble model to solve the problem of predicting the outcome of Hearthstone, a turn-based computer game, based on game state information. Excellent predictive performance of our model has been acknowledged by the second place scored in the final ranking among 188 competing teams.