This paper proposes as an element of novelty the Unified Form (UF) clustering algorithm, which treats Fuzzy C-Means (FCM) and K-Means (KM) algorithms as a single configurable algorithm. UF algorithm ...was designed to facilitate the FCM and KM algorithms software implementation by offering a solution to implement a single algorithm, which can be configured to work as FCM or KM. The second element of novelty of this paper is the Partitional Implementation of Unified Form (PIUF) algorithm, which is built upon the UF algorithm and designed to solve in an elegant manner the challenges of processing large datasets in a sequential manner and the scalability of the UF algorithm for processing datasets of any size. PIUF algorithm has the advantage of overcoming any possible hardware limitations that can occur if large volumes of data are processed (required to be stored, loaded in memory and processed by a certain specified computational system). PIUF algorithm is designed and formulated to be used on a single machine if the processed dataset is very big and it cannot be entirely loaded in the memory; at the same time it can be scaled to multiple processing nodes for reducing the processing time required to find the optimal solution. UF and PIUF algorithms are implemented and validated in BigTim platform, which is a distributed platform developed by the authors, and offers support for processing various datasets in a parallel manner but they can be implemented in any other data processing platforms. The Iris dataset is considered and next modified to obtain different datasets of different sizes in order to test the algorithms implementations in BigTim platform in different configurations. The analysis of PIUF algorithm and the comparison with FCM, KM and DBSCAN clustering algorithms are carried out using two performance indices; three performance indices are employed to evaluate the quality of the obtained clusters.
•A unified form (UF) to treat Fuzzy C-means and K-means algorithms is proposed.•UF algorithm reduces the effort required for the software implementation.•UF algorithm runs as a distributed algorithm.•UF algorithm is implemented and validated using BigTim distributed platform.•The results are analyzed and compared using several performance indices.
A big volume of new data is generated in every moment of the day by different devices and domains as social network, mobile and desktop devices, financial transaction, online websites, different ...search engines and a lot of smart home devices. The generated data is diversified and can be structured or unstructured. Clustering is the process of categorizing a dataset in groups of records that are similar and are called clusters, the grouping process being performed using a specific criterion. The K-means clustering algorithm is still popular after many years. Different versions of the K-means algorithm emerged along the time and were focused on improving the K-means algorithm by performing some preprocessing steps or by reducing the number of iterations having as a final objective the improvement of the processing time. This paper presents a way of improving the resulted clusters generated by the K-means algorithm by post processing the resulted clusters with a supervised learning algorithm. The proposed approach is focused on improving the quality of the resulting clusters and not on reducing the processing time.
Centroid Update Approach to K-Means Clustering BORLEA, I.-D.; PRECUP, R.-E.; DRAGAN, F. ...
Advances in Electrical and Computer Engineering,
01/2017, Letnik:
17, Številka:
4
Journal Article
Recenzirano
Odprti dostop
The volume and complexity of the data that is generated every day increased in the last years in an exponential manner. For processing the generated data in a quicker way the hardware capabilities ...evolved and new versions of algorithms were created recently, but the existing algorithms were improved and even optimized as well. This paper presents an improved clustering approach, based on the classical k-means algorithm, and referred to as the centroid update approach. The new centroid update approach formulated as an algorithm and included in the k-means algorithm reduces the number of iterations that are needed to perform a clustering process, leading to an alleviation of the time needed for processing a dataset. Index Terms--clustering algorithms, clustering methods, data analysis, data mining, machine learning algorithms.
The majority of present clustering algorithms are designed to work in a sequential manner processing offline data on a single machine. Different approaches were considered for parallelizing the ...existing clustering algorithms to run in a parallel manner for increasing the size of the dataset that can be processed by a computational system. This paper suggests an extended architecture of the BigTim platform such that to be capable of running clustering algorithms in a distributed manner on mobile devices. The paper presents an enhanced version of the BigTim platform, which will run clustering algorithms on a cluster of devices composed by mobile devices (mobile phones and tablets) and computers connected through a network and working together in a MapReduce manner.
Frequency Domain Design of Controllers for Lighting Process Borlea, Alexandra-Bianca; Precup, Radu-Emil; Borlea, Ioan-Daniel ...
2020 IEEE 14th International Symposium on Applied Computational Intelligence and Informatics (SACI)
Conference Proceeding
The lighting process is one of the biggest power consumption process that have place in every building, from the total energy consumption of a building 25% to 40% being allocated to the lighting ...process. Nowadays intelligent smart light bulbs based on the LED technology and having internet connection are present on the market improving the power consumption of a household. This paper proposes the frequency domain design of different types of controllers used for maintaining the light level from a room which is illuminated with smart light bulbs to a desired level, improving the comfort and reducing the power consumption by maintaining the light intensity to a desired level. The first-order plus time delay and second-order plus time delay mathematical models are used in the controllers design for lighting process. A comparison of digital simulation results is included.
Lighting process generally takes between 25%-40% of total electricity consumption in a building. In the last years the lighting systems evolved to reduce the power consumption of a building reducing ...this way the costs and environmental pollution. Nowadays smart home automation systems were developed to reduce the power consumption and improve the comfort. This paper presents and describes the lighting process of a room, and suggests the architecture of an intelligent platform. The platform is capable to maintain the light intensity in a room to a desired level using natural daylight to illuminate a room in order to reduce the power consumption.
The information stored in a database can be processed for finding patterns, group the records in classes of records using a criterion or extracting valuable information that is hidden between ...database records. The artificial intelligence domain is used to analyze big volumes of data using special algorithms designed to handle a lot of information. The time needed by a dataset analysis algorithm to process a dataset usually increases with the size of the processed dataset. Giving the fact that the hardware components have evolved in the last years, the dataset analysis algorithms can be parallelized nowadays. This paper presents a parallel implementation of the K-means clustering algorithm on a Windows based operating systems using the MapReduce approach.
Most of the clustering algorithms are designed to work as a sequential algorithm that requires all data to be present, which limits the actual implementation to run on a single machine and does not ...support horizontal scalability. This is problematic in today's context when volume of data gets larger each day and the need to process data quickly is essential. Hence, in this paper we propose a platform that allows running clustering algorithms in a distributed manner. This is achieved through splitting the data into smaller and equal partitions, and through redesigning the original clustering algorithms to allow working on a sub-set of the input data without having to interact with the processing of the rest of the input data. At the end the so-called reduce phase aggregates the partial results obtained from processing each partition and it produces the global result.
In the last years the volume of data that was generated by the mankind has increased and the complexity of data generated has also increased. Since the computers have evolved and provide more ...processing power, it is possible to carry out the real-time analysis of big volumes of data. This paper suggests the architecture of a big data processing platform called BigTim, which is able to run clustering algorithms on Windows operating systems. The platform will allow running clustering algorithms in a parallelized manner using the advantage that the entire processing power of a multi-core computer.
Evolving fuzzy models for Anti-lock Braking Systems Precup, Radu-Emil; Bojan-Dragos, Claudia-Adina; Hedrea, Elena-Lorena ...
2017 IEEE International Conference on Computational Intelligence and Virtual Environments for Measurement Systems and Applications (CIVEMSA),
2017-June
Conference Proceeding
This paper suggests evolving Takagi-Sugeno-Kang (T-S-K) fuzzy models that characterize the nonlinear dynamics phenomena occurring in the longitudinal slip of Anti-lock Braking Systems (ABSs). The ...rule bases and the parameters of the T-S-K fuzzy models are evolved by an incremental online identification algorithm (IOIA). A set of real-time experiments is conducted in order to validate the evolving T-S-K fuzzy models that describe the dynamics of the longitudinal slip in an ABS laboratory equipment setup aiming the longitudinal slip control. The experimental results prove the very good performance of the T-S-K fuzzy models in terms of fast output responses and small root mean square error values. The performance comparison with similar T-S-K fuzzy models evolved by another IOIA and three nature-inspired optimization algorithms is included.