In this paper, we analyze global climate change patterns using unsupervised machine learning techniques, specifically hierarchical clustering, which is a less known method for time series analysis. Focusing on Oujda, a Moroccan semiarid city as a case study, temperature and precipitation data from its weather station are clustered to identify distinct weather states. The primary objective is to investigate the efficiency of using hierarchical clustering on time series datasets and more specifically on climate time series to identify clusters that shows the climate changes and the shifts in transitions between meteorological seasons over time. Our study shows that this class of machine learning methods can give decent quality clustering for time series data and thus it helps discovering relevant patterns among it. Furthermore, our study is a new and additional evidence for climate change worldwide which is based on unsupervised machine learning.

This report presents a study on the paralleling of the k-nearest neighbors (KNN) algorithm using the parallel random-access machine (PRAM) model. The sequential version of the KNN algorithm is first provided, followed by the proposed PRAM algorithm corresponding to it. The paralleled version of the KNN algorithm is then implemented and its performance is compared to that of the sequential version. The results show a significant improvement in performance due to paralleling, making the paralleled KNN algorithm a viable option for high performance computing applications.

This report presents the design and development of a content-based image retrieval (CBIR) system based on color, texture, and shape, and evaluates the accuracy of the metrics cosine, Euclidean, Chebychev, and Manhattan using the Corel dataset. The implementation of the system is based on two components: feature extraction and similarity measure. For feature extraction, color, texture, and shape are extracted from each image after partitioning it on 4 sub-images. For similarity measure, the four metrics of cosine, Euclidean, Chebychev, and Manhattan are implemented and evaluated using the Corel dataset. The evaluation results show that the cosine metric achieved the highest average retrieval rate. The results of the evaluation demonstrate that the CBIR system is effective in retrieving images based on the features of color, texture, and shape.

In this paper, we are going to explore the results of a study about the conception of a rocket used to put a satellite into orbit around the Earth. This launch requires a quantity of kinetic energy called ∆V budget; it is proportionate to the height of the orbit we want to launch our satellite to. Therefore, the higher the orbit, the bigger the ∆V budget, meaning more propellant is needed. Most modern rockets are engineered to have a payload (located in the head of the rocket) and multiple other parts called “stages” in which the propellant is contained. However, the distribution of the propellant mass between the stages is a problem to optimize in order to maximize the payload to propellant mass ratio. In other words, this optimization seeks to launch a heavy satellite and put it into orbit consuming the least quantity possible of propellant. From another perspective, this optimization looks to avoid using more propellent so that the rocket to lift wouldn’t be so heavy, therefore the need for more propellant, and this makes us stuck on a quasi-infinite loop of adding mass and having to deal with it by adding mass. Optimizing such operations can help us reduce the costs of space missions, and also reduce the impact of greenhouse gas effects on the environment due to the propellant combustion. We get into a 1-dimensional analytical function, based on the Tsiolkovsky rocket equation, that associates the mass ratio of the stages, the mass ratio of the payload, and the total mass of the rocket. Once the constants are fixed, we use the optimize.minimize_scalar optimization method built-in the Python SciPy library to find the optimal configuration for every given ∆V.