EEG analysis of Parkinson's disease using time-frequency analysis and deep learning

Gene clustering based on amino acid sequence similarity has been one of the most important problems and always challenging in molecular biology. The most conventional methods are based on alignment-technique. These methods cannot identify and classify sequences, especially when the lengths of sequence are long and unequal. Therefore, in order to classify fungal hexosaminidase amino acid sequences and put them in the right taxonomical group we evaluate the feasibility of computational free alignment methods based on machine learning classifiers such as SVM, KNN, SOM and ensemble technique. The classifiers have appropriately categorized large Dikarya hexosaminidase amino acid sequences as data sets according to their taxonomical groups in two phyla named, the “Ascomycota” and the “Basidiomycota”. Two statistical methods including paired t test and PCA were used for the feature selection and reduce the dimensionality of the features, respectively. Seven classifier performance metrics, randomized complete block design, pairwise Tukey’s honestly significant difference tests and the technique for order preference by similarity to ideal solution with modified k-fold cross validation have been used as tools in order to evaluate and ranking of classifiers. In this study, the effect of training data size on the classifier performance was investigated. The results showed that the rank and the performance of classifiers were depended on the training data size. The highest obtained values for the average overall accuracy of the following training data sizes, 80, 60, 40 and 20% using KNN, KNN, ensemble and ensemble classifier were 96.96, 95.81, 94.47 and 92.47%, respectively.

Free alignment classification of dikarya fungi using some machine learning methods

Predictive modelling of the physical demands during training and competition in professional soccer players.

The aim of the smart city is to provide the technological access for the automation of city. This paper examines an audio classification application based on Machine Learning for detecting urban noise that is one of the major problems in many cities nowadays. There is an urgent need to develop an automatic urban noise detection system in enforcing public security and safe neighborhood. The urban noise detection was conducted using various Machine Learning algorithms including Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM), Support Vector Machines, Decision Tree, Random Forest, Naive Bayes with 311 dataset and urban noise dataset. Our experiments validated that CNN shows the best performance (98% accuracy) compared to other ML algorithms. The prototype of the proposed system, Smart311, was developed for automatic urban noise detection in smart cities. The noise detection was conducted on mobile devices and connected to a real-time complaint system by sending a 311 request automatically.

Smart 311 Request System with Automatic Noise Detection for Safe Neighborhood

The paper attempts to validate the effectiveness of tree classifiers to classify tabla strokes especially the ones which are overlapping in nature. It uses decision tree, ID3 and random forest as classifiers. A custom made data sets of 650 samples of 13 different tabla strokes were used for experimental purpose. 31 different features with their mean and variances were extracted for classification. Three data sets consisting of 21361, 18802 and 19543 instances respectively were used for the purpose. Validation has been done using measures like ROC curve and accuracy. The experimental results showed that all the classifiers showing excellent results with random forest outperforming the other two. The effectiveness of random forest in classifying strokes which are overlapping in nature is done by comparing the known results of that with multi-layer perceptron.

Tree-Based Classification of Tabla Strokes

The paper proposes classification of tabla strokes using multilayer feed forward artificial neural network. It uses 62 features extracted from the audio file as input units to the input layer and 13 tabla strokes as output units in the output layer. The classification has been done using dimension reduction and without using dimension reduction. The dimension reduction has been performed using Principal Component Analysis (PCA) which reduced the number of features from 62 to 28. The experiments have been performed on two sets of tabla strokes, which are played by professional tabla players, each comprises of 650 tabla strokes. The results demonstrate that correct classification of instances is more than 98 % in both the cases.

Classification of Tabla Strokes Using Neural Network

We propose a methodology for generating creative cartoon art. The system generates cartoon by looking at various existing images of cartoon characters and learning about their posture/animation style. The proposed system is creative in nature as it generates unique cartoon art by deviating from the existing styles learned by the algorithm. We build over Generative Adversarial Networks (GAN) with unsupervised learning, which have shown the ability to learn to generate novel cartoons by simulating a given distribution. The proposed model exhibits an ability to generate cartoons which are creative and novel in design. We have conducted experiments by considering around 12K Tom’s cartoon images for training purposes. The results show that the increase in number of epochs resulted in better classification accuracy. The proposed system generates the character Tom’s cartoons which are novel and we have validated the same by applying Colton’s creativity benchmark.

GANTOON: Creative Cartoons Using Generative Adversarial Network

The paper proposes a methodology to create valid music compositions using genetic algorithm. Indian percussion instrument tabla is used as a prototype for this purpose. For any given taal, the methodology could generate new compositions from randomly generated initial population of standard bols of the tabla. A unique alpha-numeric representation is used for string representation. Typical genetic operators like selection, crossover and mutation have been used, but with tailor made modifications to incorporate unique features of the instrument under study. Fitness function incorporates the concept of fuzzy string matching. Experiments were conducted using different taals and different population sizes. The computer-generated compositions have been validated by human experts for its validity and novelty.

Genetic Algorithm to Generate Music Compositions: A Case Study with Tabla

Parkinson’s disease (PD) is one of the most rapidly growing neurodegenerative diseases in the world. Due to motor symptoms, it affects the normal life of a person. There is a severe need to identify PD in its early stage to avoid it getting worse and to control its symptoms easily. The advancements in Artificial Intelligence (AI) and the Internet of Things (IoT) open up new avenues for the analysis of various data points such as the gait of a person for early-stage detection. In this paper, we propose a methodology based on the use of Long Short-Term Memory (LSTM) architecture for PD diagnosis. We have used time series analysis to find the gait patterns and deep learning techniques to extract the features and to build a classifier model. The proposed model is predicting the PD disease with 85% testing accuracy and with an F1 score of 0.90. The validation is performed using Cohen’s Kappa statistical method and obtained a score of 0.631.

Subodh Deolekar

Papers

Tree-Based Classification of Tabla Strokes

Classification of Tabla Strokes Using Neural Network

GANTOON: Creative Cartoons Using Generative Adversarial Network

Genetic Algorithm to Generate Music Compositions: A Case Study with Tabla

Early Diagnosis of Parkinson’s Disease Using LSTM: A Deep Learning Approach