Abstract This paper introduces Geometric- k -means (or $${\mathsf{G}}k$$ -means for short), a novel approach that significantly enhances the efficiency and energy economy of the widely utilized k -means algorithm, which, despite its inception over five decades ago, remains a cornerstone in machine learning applications. The essence of $${\mathsf{G}}k$$ -means lies in its active utilization of geometric principles, specifically scalar projection, to significantly accelerate the algorithm without sacrificing solution quality. This geometric strategy enables a more discerning focus on data points that are most likely to influence cluster updates, which we call as high expressive data (HE). In contrast, low expressive data (LE), does not impact clustering outcome, is effectively bypassed, leading to considerable reductions in computational overhead. Experiments spanning synthetic, real-world and high-dimensional datasets, demonstrate $${\mathsf{G}}k$$ -means is significantly better than traditional and state of the art (SOTA) $$k$$ -means variants in runtime and distance computations (DC). Moreover, $${\mathsf{G}}k$$ -means exhibits better resource efficiency, as evidenced by its reduced energy footprint, placing it as more sustainable alternative. The software code and data for our algorithm is available at https://github.com/parichit/Geometric-k-means .

This paper addresses the challenge of optimizing user grouping in Index Modulation-based Orthogonal Frequency-Division Multiple Access (IM-OFDMA) systems within dynamic and stochastic noise environments. Utilizing the eXtreme Gradient Boosting (XGBoost) machine learning algorithm, we devised a framework capable of accurately predicting the optimality of user groupings across varying Signal-to-Noise Ratio (SNR) levels. Six models corresponding to different noise conditions were created, showcasing adaptability to adjacent noise levels via distribution shift handling, thereby ensuring robust performance across a wide noise spectrum. To accurately identify the most appropriate optimality prediction model for dynamic environments, we introduced a specialized model for precisely estimating the system’s internal noise power. The accuracy of this model is crucial for the selection process and was significantly enhanced by implementing sequential Bayesian updating, facilitating a more precise estimation of internal noise power. Following this, we introduced a provisional optimization algorithm designed to refine user groupings within dynamic IM-OFDMA systems. Simulation results highlight the algorithm’s effectiveness in markedly improving system performance, evidenced by a significant decrease in bitrate errors. These findings illuminate the significant potential of applying machine learning strategies to wireless communication systems, providing insightful contributions to the enhancement of IM-OFDMA systems in practical settings.

This paper introduces a novel method for time series forecasting using de Bruijn Graphs (dBGs) to represent discretized time series data. Our approach involves (1) encoding time series as a dBG, (2) applying both novel and existing graph encoding algorithms (like struct2vec) to extract features from dBG, and (3) integrating these features into the TimesNet model to enhance short-term univariate forecasting accuracy. Empirical results on the M4 datasets show that our method preserves the dynamics of the time series while improving forecasting performance across various datasets.

This paper introduces$p$-ClustVal, a novel data transformation technique inspired by p-adic number theory that significantly enhances cluster discernibility in genomics data, specifically Single Cell RNA Sequencing (scRNASeq). By lever-aging p-adic-valuation,$p$-ClustVal integrates with and augments widely used clustering algorithms and dimension reduction techniques, amplifying their effectiveness in discovering meaningful structure from data. The transformation uses a data-centric heuristic to determine optimal parameters, without relying on ground truth labels, making it more user-friendly.$p$-ClustVal reduces overlap between clusters by employing alternate metric spaces inspired by p-adic-valuation, a significant shift from conventional methods. Our comprehensive evaluation spanning 30 experiments and over 1200 observations, shows that$p$-ClustVal improves performance in 91% of cases, and boosts the performance of classical and state of the art (SOTA) methods. This work contributes to data analytics and genomics by introducing a unique data transformation approach, enhancing downstream clustering algorithms, and providing empirical evidence of$p$-ClustVal's efficacy.

Fantasy Sports has a current market size of ${\$}$27B and is expected to grow more than ${\$}$84B in less than a decade. The intent is to create virtual teams that somehow reflect what would happen if the constituent players actually played in a team. Using individual player and team statistics, models can be trained to predict an outcome. But fans are left wanting more. To achieve a more realistic outcome, aspects of what makes live teams win need to be included: (1) transforming player statistics to reflect their relative importance with respect to a player position; (2) team chemistry (TC). In this work, we show a novel characterization of relative position statistics and a new description of TC. Drawn from the NBA’s API, we form a data set to determine whether a fantasy team makes the playoffs using almost two dozen features, including TC. Various Machine Learning models are trained on this data and the best-performing model is offered to the users through a web service. Users can not only inspect fantasy teams and their TC but can also simulate their match-ups with existing 2023 NBA teams and utilize performance visualizations to help improve their team creation process. Our web service can be accessed at https://dalkilic.luddy.indiana.edu/fantasyleague/, and the source code can be found at https://github.com/gany-15/nbafan.

BACKGROUND: In recent years, the introduction of single-cell RNA sequencing (scRNA-seq) has enabled the analysis of a cell's transcriptome at an unprecedented granularity and processing speed. The experimental outcome of applying this technology is a [Formula: see text] matrix containing aggregated mRNA expression counts of M genes and N cell samples. From this matrix, scientists can study how cell protein synthesis changes in response to various factors, for example, disease versus non-disease states in response to a treatment protocol. This technology's critical challenge is detecting and accurately recording lowly expressed genes. As a result, low expression levels tend to be missed and recorded as zero - an event known as dropout. This makes the lowly expressed genes indistinguishable from true zero expression and different than the low expression present in cells of the same type. This issue makes any subsequent downstream analysis difficult. RESULTS: To address this problem, we propose an approach to measure cell similarity using consensus clustering and demonstrate an effective and efficient algorithm that takes advantage of this new similarity measure to impute the most probable dropout events in the scRNA-seq datasets. We demonstrate that our approach exceeds the performance of existing imputation approaches while introducing the least amount of new noise as measured by clustering performance characteristics on datasets with known cell identities. CONCLUSIONS: ccImpute is an effective algorithm to correct for dropout events and thus improve downstream analysis of scRNA-seq data. ccImpute is implemented in R and is available at https://github.com/khazum/ccImpute .

Using a large volume of bus data in the form of GPS coordinates (over 100 million data points) and automated passenger count data (over 1 million data points) we have developed (1) a system of analysis and prediction of future public transportation demand (2) a new model that uses concepts specific to college campuses that maximizes passenger satisfaction. Using these concepts we improve service of a model college public transportation service and more specifically the Indiana University Campus Bus Service (IUCBS).

Without question, astronomy is about Big Data and clustering is a very common task over astronomy domain. The expectation-maximization algorithm is among the top 10 data mining algorithms used in scientific and industrial applications, however, we observe that astronomical community does not make use of it as a clustering algorithm. In this work, we cluster $\sim$ 1M stellar objects (simulated Galactic spectral data) via the traditional expectation-maximization algorithm for clustering (EM-T) and our extended EM-T algorithm that we call EM* and present the experimental results.

Iterative machine learning algorithms, i.e., k-means (KM), expectation maximization (EM), become overwhelmed with big data since all data points are being continually and indiscriminately visited while a cost is being minimized. In this work, we demonstrate (1) an optimization approach to reduce training run-time complexity of iterative machine learning algorithms and (2) implementation of this framework over KM algorithm. We call this extended KM algorithm, KM*. The experimental results show that KM* outperforms KM over big real world and synthetic data sets. Lastly, we demonstrate the theoretical elements of our work.

Stellar data, only a few years ago, measured in the .1M of objects. Now, sets are routinely 1M. With the launch of ESA's Gaia in 2013, we expect 1000M stellar objects measured more precisely and with more measurements. Without question, astronomy is about Big Data and clustering is a very common task over astronomy domain. The expectation-maximization algorithm is among the top 10 data mining algorithms used in scientific and industrial applications, however, we observe that astronomical community does not make use of it as a clustering algorithm. In this work, we cluster ~ 1M stellar objects (simulated Galactic spectral data) via the traditional expectation-maximization algorithm for clustering (EM-T) and our extended EM-T algorithm that we call EM* and present the experimental results.

Existing data mining techniques, more particularly iterative learning algorithms, become overwhelmed with big data. While parallelism is an obvious and, usually, necessary strategy, we observe that both (1) continually revisiting data and (2) visiting all data are two of the most prominent problems especially for iterative, unsupervised algorithms like Expectation Maximization algorithm for clustering (EM-T). Our strategy is to embed EM-T into a non-linear hierarchical data structure(heap) that allows us to (1) separate data that needs to be revisited from data that does not and (2) narrow the iteration toward the data that is more difficult to cluster. We call this extended EM-T, EM*. We show our EM* algorithm outperform EM-T algorithm over large real world and synthetic data sets. We lastly conclude with some theoretic underpinnings that explain why EM* is successful.

Random forests have been used as effective models to tackle a number of classification and regression problems. In this paper, we present a new type of Random Forests (RFs) called Red(uced)-RF that adopts a new voting mechanism called Priority Vote Weighting (PV) and a new dynamic data reduction principle which improve accuracy and execution time compared to Breiman's conventional RF. Red-RF also shows that the strength of a random forest can increase without noticeably increasing correlation between the trees. We then compare performance of Red-RF, 9 new RF variants and Breiman's RF in eight experiments that involve classification problems with datasets of different sizes.