In the iron ore mining fraternity, in order to achieve the desired quality in the froth flotation processing plant, stakeholders rely on conventional laboratory test technique which usually takes more than two hours to ascertain the two variables of interest. Such a substantial dead time makes it difficult to put the inherent stochastic nature of the plant system in steady-state. Thus, the present study aims to evaluate the feasibility of using machine learning algorithms to predict the percentage of silica concentrate (SiO2) in the froth flotation processing plant in real-time. The predictive model has been constructed using iron ore mining froth flotation system dataset obtain from Kaggle. Different feature selection methods including Random Forest and backward elimination technique were applied to the dataset to extract significant features. The selected features were then used in Multiple Linear Regression, Random Forest and Artificial Neural Network models and the prediction accuracy of all the models have been evaluated and compared with each other. The results show that Artificial Neural Network has the ability to generalize better and predictions were off by 0.38% mean square error (mse) on average, which is significant considering that the SiO2 range from 0.77%- 5.53% -( mse 1.1%) . These results have been obtained within real-time processing of 12s in the worst case scenario on an Inter i7 hardware. The experimental results also suggest that reagents variables have the most significant influence in SiO2 prediction and less important variable is the Flotation Column.02.air.Flow. The experiments results have also indicated a promising prospect for both the Multiple Linear Regression and Random Forest models in the field of SiO2 prediction in iron ore mining froth flotation system in general. Meanwhile, this study provides management, metallurgists and operators with a better choice for SiO2 prediction in real-time per the accuracy demand as opposed to the long dead time laboratory test analysis causing incessant loss of iron ore discharged to tailings.

i Abstract In this work, we describe a novel methodology to analyse the price elasticity of demand. This method combines an artificial neural network that serves as the model of the behaviour of the customers and a subsequent simulation based on this model. We present the validation of our approach using a real-world dataset obtained from an e-commerce retailer and demonstrate its advantages, notably the ability to estimate the elasticity in distinct price points and the inclusion of the complete pricing situations (not only product's own price). JEL Classification C45, C44, C15, D12 Keywords price elasticity of demand, artificial neural net- work, agent-based model Title A simulation based analysis of price elasticity of demand Author's e-mail Supervisor's e-mail

A method that uses machine learning to detect and localize damage in railway bridges under various environmental conditions is proposed and validated in this work. The developed algorithm uses vertical and lateral deck accelerations as damage- sensitive features. Indeed, an Artificial Neural Network (ANN) is trained to predict deck accelerations in undamaged condition given: previous vibration data, air temperature and characteristics of the train crossing the bridge (speed, load position and load magnitude). After an appropriate training period, the comparison between ANN-predicted and measured accelerations allows to compute prediction errors. A Gaussian Process is then used to stochastically characterize prediction errors in undamaged conditions using train speed as independent variable. Recorded vibration data leading to abnormal prediction errors are flagged as damage. The method is validated both on a simple numerical example and on data recorded on a real structure. In the latter case, an appropriate algorithm was developed with the aim of extracting vehicles characteristics from the acceleration time histories. Together with this part of the algorithm for the pre-processing of recorded accelerations, the novelty of the developed method is the addition of air temperature to the input. It allows separating between structure responses that can be flagged as damage from those only affected by environmental conditions.

While global efforts are made to reduce the emission of greenhouse gases and move towards a more sustainable society, the global energy demand is continuing to increase. Building energy consumption represents 20-40% of the world's total energy use, and Heating, Ventilation, and Air Conditioning (HVAC) answer for around 50% of this amount. Only a small share of the European Union's building stock is considered to be energy efficient, and many of these buildings will continue to operate until the year 2050 and on-wards. The main objective of this thesis was to benchmark the economic and environmental implications of increasing building heating efficiency. To answer the framed research questions, an experimental study was carried out. In the study, a machine learning based solution was constructed and then implemented in a multi-tenant building for 24 days. Using an Artificial Neural Network a new heating curve was predicted, based on historical data from the building. The post-experimental data was then analyzed using STATA as statistical software tool. The results show that the new heating curve was able to reduce the heating system supply temperature by 1.9°C, with a decrease in average indoor temperature of 0.097°C. The decrease in supply temperature resulted in a reduction of energy expenditure by approximately 10%. Using the new building specific heating curve, yearly cost reductions of almost 11,700SEK could be achieved. Furthermore, the increased efficiency was able to reduce CO2 emissions by 127,5kg yearly. This results helps shed light on the general weaknesses in building heating systems out there today, and shows that there is great potential of reducing building energy consumption in cost effective ways. Although the implemented solution might not be generally applicable for all building owners out there, it should act as an eye opener for building owners and help motivate them into assessing their building operation and start looking into new technologies. Moreover, the study provides legible incentives for both building owners and the society to further work together towards a more efficient and sustainable society.

Maintaining the electricity balance between supply and demand is a challenge for electricity suppliers. If there is an under or overproduction, it entails financial costs and affects consumers and the climate. To better understand how to maintain the balance, can the suppliers use short-term forecasts of electricity load. Hence it is of paramount importance that the forecasts are reliable and of high accuracy. Studies show that time series modeling moves towards more data-driven methods, such as Artificial Neural Networks due to their ability to extract complex relationships and flexibility. This study evaluates the performance of a multivariate Deep Autoregressive Neural Network (DeepAR) in ashort-term forecasting scenario of electricity load, with forecasted weather parameters as exogenous covariates. This thesis’s goal is twofold: to test the performance in terms of evaluation metrics of day-ahead forecasts in exogenous covariates’ presence and examine the robustness when exposing DeepAR to deviations in input data. We perform feature selection on given covariates to identify and extract relevant parameters to facilitate the training process and implement a feature importance algorithm to examine which parameters the model considers essential. To test the robustness, we simulate two cases. In the first case, we introduce Quarantine periods, which mask data prior to the forecast range, and the second case introduces an artificial outlier. An exploratory analysis displays significant annual characteristic differences between seasons, therefore do we use two test sets, one in winter and one in summer. The result shows that DeepAR is robust against potential deviations in input data and that DeepAR surpassed both benchmark models in all of the tested scenarios. In the ideal test scenario where weather parametershad the most significant impact (winter), do DeepAR achieve a Normalized Deviation(ND) of 2.5%, compared to the second-best model, with an ND of 4.4%

Under de senaste åren fick maskininlärning mer och mer popularitet i samhället. Den implementeras i stor utsträckning inom många datavetenskapliga områden, t.ex. igenkänning av tal, video, objekt, sentimentanalys osv. Dessutom genererar moderna datorsystem och program stora filer med loggdata under deras körning och användning. Dessa loggfiler innehåller vanligtvis enorma mängder data, vilket leder till svårigheter att bearbeta all data manuellt. Således är användning av maskininlärningstekniker vid analys av loggdata för detektering av anomalibeteende av stort intresse för att uppnå skalbar underhåll av systemen. Syftet med detta arbete var att undersöka tillgängliga framträdande metoder för att implementera maskininlärning för upptäckning av loggfel och utvärdera en av dessa metoder. Uppsatsen fokuserade på att utvärdera DeepLog artificiella neurala nätverk som innehåller Long short-term memory algoritm. Utvärderingen omfattade mätning av den exekveringstid som behövdes och vilken precision, återkallande, noggrannhet och F1-index uppnåddes med modellen för maskininlärningsfelsdetektering vid användning av två olika loggdatamängder, en från OpenStack och en annan från Hadoop Distributed File System. Resultaten visade att DeepLog presterade bättre när man använde OpenStack-datamängd genom att uppnå höga resultat för alla index, särskilt recallsindex på cirka 90% som minimerade falska negativa förutsägelser, vilket är viktigt vid loggfelsdetektering. När DeepLog användes med HDFS-datamängd förbättrades körningstiden något men noggrannheten och recall av modellen tappades. Framtida arbete inkluderar att försöka och testa modellen med andra loggdatamängder eller andra ML-modeller för upptäckning av loggfel. / During the last years machine learning was gaining more and more popularity in the society. It is widely implemented in many fields of computer science, e.g. recognition of speech, video, objects, sentiment analysis, etc. Additionally, modern computer systems and programs generate large files with log data through their execution. These log files contain usually immense amount of data, which is a struggle for processing it manually. Thus, using machine learning techniques in the analysis of log data for detection of anomaly behavior is of a high interest for achieving scalable maintaining of the systems. The purpose of this work was to look into available prominent approaches of implementing machine learning for log fault detection and evaluate one of them. The paper focused on evaluating DeepLog artificial neural network that incorporates Long short-term memory. The evaluation included measuring the execution time needed and what precision, recall, accuracy and F1-index were achieved by the machine learning fault detection model when using two different log datasets, one from OpenStack and another from Hadoop Distributed File System. The results showed that DeepLog performed better when using OpenStack dataset by achieving high results for all indexes, especially the recall index of around 90%, minimizing the false negative predictions, which is important in the log fault detection. When using DeepLog with HDFS dataset the execution time was slightly improved but the accuracy and recall of the model were dropped. Future works includes trying another log datasets or ML models for log fault detection.

The quality of a wine is important for the consumers as well as the wine industry. The traditional (expert) way of measuring wine quality is time-consuming. Nowadays, machine learning models are important tools to replace human tasks. In this case, there are several features to predict the wine quality but the entire features will not be relevant for better prediction. So, our thesis work is focusing on what wine features are important to get the promising result. For the purposeof classification model and evaluation of the relevant features, we used three algorithms namely support vector machine (SVM), naïve Bayes (NB), and artificial neural network (ANN). In this study, we used two wine quality datasets red wine and white wine. To evaluate the feature importance we used the Pearson coefficient correlation and performance measurement matrices such as accuracy, recall, precision, and f1 score for comparison of the machine learning algorithm. A grid search algorithm was applied to improve the model accuracy. Finally, we achieved the artificial neural network (ANN) algorithm has better prediction results than the Support Vector Machine (SVM) algorithm and the Naïve Bayes (NB) algorithm for both red wine and white wine datasets.

Slay the Spire is a complex deck-building and roguelike game with many possibilities of improving players ability to win. An important part of Slay the Spire is choosing a path that makes the players character as successful as possible. In this study we show that machine learning can help players pick better paths by creating an Artificial Neural Network (ANN) that predicts the most successful path of all available paths, we also discuss what makes a path successful. This study performed two experiments, one user study and one simulation experiment, with the intention of evaluating the created ANN and analysing what makes paths successful. Through the user study this paper shows that the ANN was effective at predicting paths, outperforming all other human players who played normally in all three cases. This study concludes that machine learning can be used effectively to help make pathing decisions in Slay the Spire. Furthermore the study proves the importance of the room types ’Elite’ and ’Campfires’ through the simulation experiment, user study and analysis of data from previous playthroughs.

L’objectif de ces travaux de doctorat est de développer un modèle prédictif capable de prévoir correctement les concentrations en polluants du jour pour le lendemain en Corse. Nous nous sommes intéressés aux PM10 et à l’ozone, les deux polluants les plus problématiques sur l’île. Le modèle devait correspondre aux contraintes d’un usage opérationnel au sein d’une petite structure, comme Qualitair Corse, l’association locale de surveillance de la qualité de l’air.La prévision a été réalisée à l’aide de réseaux de neurones artificiels. Ces modèles statistiques offrent une grande précision tout en nécessitant peu de ressources informatiques. Nous avons choisi le Perceptron MultiCouche (PMC), avec en entrée à la fois des mesures de polluants, des mesures météorologiques, et des sorties de modèles de chimie-transport (CHIMERE via la plate-forme AIRES) et de modèles météorologiques (AROME).La configuration des PMC a été optimisée avant leur apprentissage automatique, en conformité avec le principe de parcimonie. Pour en améliorer les performances, une étude de sélection de variables a été au préalable menée. Nous avons comparé l’usage d’algorithmes génétiques, de recuits simulés et d’analyse en composantes principales afin d’optimiser le choix des variables d’entrées. L’élagage du PMC a été également mis en œuvre.Nous avons ensuite proposé un nouveau type de modèle hybride, combinaison d’un classifieur et de plusieurs PMC, chacun spécialisé sur un régime météorologique particulier. Ces modèles, qui demandent un large historique de données d’apprentissage, permettent d’améliorer la prévision des valeurs extrêmes et rares, correspondant aux pics de pollution. La classification non-supervisée a été menée avec des cartes auto-organisatrices couplées à l’algorithme des k-means, ainsi que par classification hiérarchique ascendante. L’analyse de sensibilité à été menée grâce à l’usage de courbes ROC.Afin de gérer les jeux de données utilisés, de mener les expérimentations de manière rigoureuse et de créer les modèles destinés à l’usage opérationnel, nous avons développé l’application « Aria Base », fonctionnant sous Matlab à l’aide de la Neural Network Toolbox.Nous avons également développé l’application « Aria Web » destinée à l’usage quotidien à Qualitair Corse. Elle est capable de mener automatiquement les prévisions par PMC et de synthétiser les différentes informations qui aident la prévision rendues disponibles sur internet par d’autres organismes. / The objective of this doctoral work is to develop a forecasting model able to correctly predict next day pollutant concentrations in Corsica. We focused on PM10 and ozone, the two most problematic pollutants in the island. The model had to correspond to the constraints of an operational use in a small structure like Qualitair Corse, the local air quality monitoring association.The prediction was performed using artificial neural networks. These statistical models offer a great precision while requiring few computing resources. We chose the MultiLayer Perceptron (MLP), with input data coming from pollutants measurements, meteorological measurements, chemical transport model (CHIMERE via AIRES platform) and numerical weather prediction model (AROME).The configuration of the MLP was optimized prior to machine learning, in accordance with the principle of parsimony. To improve forecasting performances, we led a feature selection study. We compared the use of genetic algorithms, simulated annealing and principal component analysis to optimize the choice of input variables. The pruning of the MLP was also implemented.Then we proposed a new type of hybrid model, combination of a classification model and various MLPs, each specialized on a specific weather pattern. These models, which need large learning datasets, allow an improvement of the forecasting for extreme and rare values, corresponding to pollution peaks. We led unsupervised classification with self organizing maps coupled with k-means algorithm, and with hierarchical ascendant classification. Sensitivity analysis was led with ROC curves.We developed the application “Aria Base” running with Matlab and its Neural Network Toolbox, able to manage our datasets, to lead rigorously the experiments and to create operational models.We also developed the application “Aria Web” to be used daily by Qualitair Corse. It is able to lead automatically the prevision with MLP, and to synthesize forecasting information provided by other organizations and available on the Internet.

No description available.