Koushik Ghosh

An affinity-based new local distance function and similarity measure for kNN algorithm

In this paper, we propose a modified version of the k-nearest neighbor (kNN) algorithm. We first introduce a new affinity function for distance measure between a test point and a training point which is an approach based on local learning. A new similarity function using this affinity function is proposed next for the classification of the test patterns. The widely used convention of k, i.e., k=[@/N] is employed, where N is the number of data used for training purpose. The proposed modified kNN algorithm is applied on fifteen numerical datasets from the UCI machine learning data repository. Both 5-fold and 10-fold cross-validations are used. The average classification accuracy, obtained from our method is found to exceed some well-known clustering algorithms.

Outlier detection using neighborhood rank difference

Use of neighborhood rank-difference for outlier score.Dynamic (dataset specific) k for construction of influence/decision space.High rank-power for both synthetic and real datasets. Display Omitted Presence of outliers critically affects many pattern classification tasks. In this paper, we propose a novel dynamic outlier detection method based on neighborhood rank difference. In particular, reverse and the forward nearest neighbor rank difference is employed to capture the variations in densities of a test point with respect to various training points. In the first step of our method, we determine the influence space for a given dataset. A score for outlierness is proposed in the second step using the rank difference as well as the absolute density within this influence space. Experiments on synthetic and some UCI machine learning repository datasets clearly indicate the supremacy of our method over some recently published approaches.

Search for periodicities of the solar irradiance data from the Earth Radiation Budget Satellite (ERBS) using the periodogram method

We have analyzed the solar irradiance data from the Earth Radiation Budget Satellite (ERBS) during the time period from 1984 October 15 to 2003 October 15. By first filtering the data by Simple Exponential Smoothing, we have applied the periodogram method to the processed data in order to search for its time variation. The study exhibits multi-periodicities on these data around 110, 118, 574 and 740d with very high confidence levels (more than 99%). These periods are significantly similar to the periods of other solar activities which may suggest that solar irradiance may be associated with other solar activities.

Granger Causality Driven AHP for Feature Weighted kNN

The kNN algorithm remains a popular choice for pattern classification till date due to its non-parametric nature, easy implementation and the fact that its classification error is bounded by twice the Bayes error. In this paper, we show that the performance of the kNN classifier improves significantly from the use of (training) class-wise group-statistics based two criteria during pairwise comparison of features in a given dataset. Granger causality is employed to assign preferences to each criteria. Analytic Hierarchy Process (AHP) is applied to obtain weights for different features from the two criteria and their preferences. Finally, these weights are used to build a weighted distance function for the kNN classification. Comprehensive experimentation on fifteen benchmark datasets of the UCI Machine Learning Repository clearly reveals the supremacy of the proposed Granger causality driven AHP induced kNN algorithm over the kNN method with many different distance metrics, and, with various feature selection strategies. In addition, the proposed method is also shown to perform well on high-dimensional face and hand-writing recognition datasets. HighlightsFeature weighting for kNN by a multi-criteria based decision analysis tool called AHP.Automated weight assignment in criteria matrix of AHP using group-statistics.Criteria preference selection in AHP with Granger Causality.Superior classification performance over kNN with many other feature weighting/selection methods.

A search for latent periodicities of irregular time series of total solar irradiance

In this paper we have analyzed the total solar irradiance signal obtained from Earth Radiation Budget Satellite. After denoising this signal using simple exponential smoothing we experimented with this filtered signal to study its nature. Fractal dimension analysis has been applied on it and the obtained value of fractal dimension demonstrates anti-persistent behaviour (short memory process) which may imply multi-periodic phenomenon. To sort out the possible periods with higher confidences we have applied Ferraz-Mellos method of Date Compensated Discrete Fourier Transform as well as method of Periodogram which are very much useful for the present time series with uneven gaps. We have found certain periodicities from these two methods with confidence levels all higher than 95%.

kNN Classification with an Outlier Informative Distance Measure

Classification accuracy of the kNN algorithm is found to be adversely affected by the presence of outliers in the experimental datasets. An outlier score based on rank difference can be assigned to the points in these datasets by taking into consideration the distance and density of their local neighborhood points. In the present work, we introduce a generalized outlier informative distance measure where a factor based on the above score is used to modulate any potential distance function. Properties of the new outlier informative distance measure are presented. Experiments on several numeric datasets in the UCI machine learning repository clearly reveal the effectiveness of the proposed formulation.

Scaling and stationarity analysis of discrete time signal of Solar Radio flux

In the present work, we have considered the daily signal of Solar Radio flux of 2800 Hz recorded daily by radio telescopes near Ottawa (operated during 14th February, 1947-31st May, 1991) and Penticton, British Columbia, Dominion Radio Astrophysical Observatory (operating since 1st June, 1991) during the period from 29th October, 1972 to 28th February, 2013. We have applied FIR nonlinear phase filter on the present discrete signal to denoise it. The memory of this denoising signal has been analyzed and it indicates short memory trend (anti-persistent) hidden in this spectrum. Then we have examined whether the signal possesses stationary properties. On the basis of autocorrelation analysis, quasi or partly stationary nature of the spectrum has been revealed.

Profile and practices of Indian premier institutes compared to global standards on the basis of QS World Universities Ranking 2013–14

Not a single Indian university is among the Worlds top 200, according to a new global ranking. The QS World Universities Ranking published has miserable news for Indias education system. Around 11 Indian institutes/universities feature in the top 800 of the global lists with the highest ranking going to IIT Delhi which is passed 222 in the list. Two other made it to the top 300 are IIT Bombay and IIT Kanpur. The story was similar in the latest Times Higher Education World Universities ranking with no Indian institution being ranked among the top 200. A country of Indias rich intellectual history, vast size and growing economic power needs at least some world class universities that can compete with the very best universities in the world. This paper tries to examine the average score difference between Indian top institutes with the global top universities on the basis of QS database and tries to explore the reasons for such differences. This paper also tries to show the level of competitiveness amongst Indian top institutes and adherence to global practices for attaining global standards to catch up the top universities of the world and revive Indias lost glory.

Short-term prediction of Forbush decrease indices with Adaptive Neuro-Fuzzy Inference Systems

Forbush decrease is a rapid decrease in the observed galactic cosmic ray intensity pattern followed by a coronal mass ejection. In the present paper we have analyzed the daily sampled Forbush decrease indices from January, 1967 to December, 2003 generated in IZMIRAN, Russia. Prediction of this astrophysical parameter has been performed using Takagi-Sugeno type Adaptive Neuro-Fuzzy Inference Systems (ANFIS) that allows reduction both in statistical and systematic uncertainties. The visual observation based on the graphical comparison between observed and predicted values and the qualitative performance assessment of the model indicates that this method can be used effectively for minimum Forbush decrease indices predictions. In this work, bell-shaped Gauss type membership functions have been found suitable and Hybrid Learning Algorithm method has been used for the optimization. To judge the predictive capability of the developed methodology, based on ANFIS model, the performance indicators show that root mean square error value is 0.002505 for training and 0.002502 for testing period. So, it may establish that the Forbush effect as a storm in cosmic rays is interestingly very much predictable in short senses.