2024/09/11 by Maxime Kawawa-Beaudan, Kawawa-Beaudan, Maxime, Srijan Sood +9 · 1 citation
Computer Science · #Algorithms and Data Compression #Anomaly Detection Techniques and Applications #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.2409.07619
openalex publication_date 2024/09/11 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
We present a lightweight approach to sequence classification using Ensemble\nMethods for Hidden Markov Models (HMMs). HMMs offer significant advantages in\nscenarios with imbalanced or smaller datasets due to their simplicity,\ninterpretability, and efficiency. These models are particularly effective in\ndomains such as finance and biology, where traditional methods struggle with\nhigh feature dimensionality and varied sequence lengths. Our ensemble-based\nscoring method enables the comparison of sequences of any length and improves\nperformance on imbalanced datasets.\n This study focuses on the binary classification problem, particularly in\nscenarios with data imbalance, where the negative class is the majority (e.g.,\nnormal data) and the positive class is the minority (e.g., anomalous data),\noften with extreme distribution skews. We propose a novel training approach for\nHMM Ensembles that generalizes to multi-class problems and supports\nclassification and anomaly detection. Our method fits class-specific groups of\ndiverse models using random data subsets, and compares likelihoods across\nclasses to produce composite scores, achieving high average precisions and\nAUCs.\n In addition, we compare our approach with neural network-based methods such\nas Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks\n(LSTMs), highlighting the efficiency and robustness of HMMs in data-scarce\nenvironments. Motivated by real-world use cases, our method demonstrates robust\nperformance across various benchmarks, offering a flexible framework for\ndiverse applications.\n