Koray Açıcı

Integrating Features for Accelerometer-based Activity Recognition

Activity recognition is the problem of predicting the current action of a person through the motion sensors worn on the body. The problem is usually approached as a supervised classification task where a discriminative model is learned from known samples and a new query is assigned to a known activity label using learned model. The challenging issue here is how to feed this classifier with a fixed number of features where the real input is a raw signal of varying length. In this study, we consider three possible feature sets, namely time-domain, frequency domain and wavelet-domain statistics, and their combinations to represent motion signal obtained from accelerometer reads worn in chest through a mobile phone. In addition to a systematic comparison of these feature sets, we also provide a comprehensive evaluation of some preprocessing steps such as filtering and feature selection. The results determine that feeding a random forest classifier with an ensemble selection of most relevant time-domain and frequency-domain features extracted from raw data can provide the highest accuracy in a real dataset.

Retrieving relevant experiments: The case of microRNA microarrays.

Content-based retrieval of biological experiments in large public repositories is a recent challenge in computational biology and bioinformatics. The task is, in general, to search in a database using a query-by-example without any experimental meta-data annotation. Here, we consider a more specific problem that seeks a solution for retrieving relevant microRNA experiments from microarray repositories. A computational framework is proposed with this objective. The framework adapts a normal-uniform mixture model for identifying differentially expressed microRNAs in microarray profiling experiments. A rank-based thresholding scheme is offered to binarize real-valued experiment fingerprints based on differential expression. An effective similarity metric is introduced to compare categorical fingerprints, which in turn infers the relevance between two experiments. Two different views of experimental relevance are evaluated, one for disease association and another for embryonic germ layer, to discern the retrieval ability of the proposed model. To the best of our knowledge, the experiment retrieval task is investigated for the first time in the context of microRNA microarrays.

A Random Forest Method to Detect Parkinson’s Disease via Gait Analysis

Remote care and telemonitoring have become essential component of current geriatric medicine. Intelligent use of wireless sensors is a major issue in relevant computational studies to realize these concepts in practice. While there has been a growing interest in recognizing daily activities of patients through wearable sensors, the efforts towards utilizing the streaming data from these sensors for clinical practices are limited. Here, we present a practical application of clinical data mining from wearable sensors with a particular objective of diagnosing Parkinson’s Disease from gait analysis through a sets of ground reaction force (GRF) sensors worn under the foots. We introduce a supervised learning method based on Random Forests that analyze the multi-sensor data to classify the person wearing these sensors. We offer to extract a set of time-domain and frequency-domain features that would be effective in distinguishing normal and diseased people from their gait signals. The experimental results on a benchmark dataset have shown that proposed method can significantly outperform the previous methods reported in the literature.

Texture of Activities: Exploiting Local Binary Patterns for Accelerometer Data Analysis

Recognition of activities through wearable sensors such as accelerometers is a recent challenge in pervasive and ubiquitous computing. The problem is often considered as a classification task where a set of descriptive features are extracted from input signal to feed a machine learning classifier. A major issue ignored so far in these studies is the incorporation of locally embedded features that could indeed be informative in describing the main activity performed by the individual being experimented. To close this gap, we offer here adapting Local Binary Pattern (LBP) approach, which is frequently used in identifying textures in images, in one dimensional space of accelerometer data. To this end, we exploit the histogram of LPB found in each axes of input accelerometer signal as a feature set to feed a k-Nearest Neighbor classifier. The experiments on a benchmark dataset have shown that the proposed method can outperform some previous methods.

Information retrieval in metal music sub-genres

Digital music platforms use meta-data based information retrieval systems for offering songs to users for their own taste of music. According to this system, songs that are labeled by other users are compared to songs that user listened and similar labeled songs are retrived in the process. In this situtation, information retrieval is independent from song content and subjective. To achieve objectivity, content based information retrieval systems are needed. In this study, a content-based music retrieval system based on one dimensional local binary pattern features which are extracted from audio data is proposed. Instead of retrieving different music genres, retrieval is applied on metal music sub-genres which have not been studied before and results are reported.

Comparison of similarity metrics in microarray experiment retrieval

Content-based retrieval of biological experiments is a recent challenge in bioinformatics. The task is to search in a database using a query-by-example without any meta-data annotation. In this study, for retrieving relevant microRNA experiments from microarray repositories, performance evaluation of known similarity metrics was conducted to compare experiment fingerprints. It was shown that Spearman correlation coefficient outperformed others by comparison on real datasets. This result shows that ranks of fingerprint values are more important than the exact values in experiment fingerprint.

Inferring microarray relevance by enrichment of chemotherapy resistance-based microRNA sets

Inferring relevance between microarray experiments stored in a gene expression repository is a helpful practice for biological data mining and information retrieval studies. In this study, we propose a knowledge-based approach for representing microarray experiment content to be used in such studies. The representation scheme is specifically designed for inferring a disease-associated relevance of microRNA experiments. A group of annotated microRNA sets based on their chemotherapy resistance are used for a statistical enrichment analysis over observed expression data. A query experiment is then represented by a single dimensional vector of these enrichment statistics, instead of raw expression data. According to the results, new representation scheme can provide a better retrieval performance than traditional differential expression-based representation.

Sensitive prediction of bacterial type IV effectors

Diverse human pathogens secret effector proteins into host cells via the type IV secretion system (T4SS). Effector proteins are important elements in the interaction between bacteria and hosts. Computational methods for T4SS effector prediction will be of great value. This paper introduces five types of feature representation schemes for prediction of effectors from sequence namely, amino acid composition, dipeptide composition, three-peptide composition, BLAST similarity scores and pseudo amino acid composition. SVM, k-NN, Naïve Bayes and Fisher LDA classification methods were performed in a newly established the dataset to predict T4SS effectors with using generated features. The experimental results indicate that classification methods we used are useful to discriminate IVA and IVB effectors with positive rates 83,3%, 96,5% respectively. The overall accuracy of 95.5% shows that the present method is accurate for distinguishing the T4SS effector in unidentified sequences.