In this research, we apply data mining techniques to a nationally-representative expenditure data from the US to predict very high-cost patients in the top 5 cost percentiles, among the general population. Samples are derived from the Medical Expenditure Panel Survey’s Household Component data for 2006-2008 including 98,175 records. After pre-processing, partitioning and balancing the data, the final MEPS dataset with 31,704 records is modeled by Decision Trees (including C5.0 and CHAID), Neural Networks. Multiple predictive models are built and their performances are analyzed using various measures including correctness accuracy, G-mean, and Area under ROC Curve. We conclude that the CHAID tree returns the best G-mean and AUC measures for top performing predictive models ranging from 76% to 85%, and 0.812 to 0.942 units, respectively. Among a primary set of 66 attributes, the best predictors to estimate the top 5% high-cost population include individual’s overall health perception, history of blood cholesterol check, history of physical/sensory/mental limitations, age, and history of colonic prevention measures. It is worthy to note that we do not consider number of visits to care providers as a predictor since it has a high correlation with the expenditure, and does not offer a new insight to the data (i.e. it is a trivial predictor). We predict high-cost patients without knowing how many times the patient was visited by doctors or hospitalized. Consequently, the results from this study can be used by policy makers, health planners, and insurers to plan and improve delivery of health services.

In this research, we apply data mining techniques to a nationally-representative expenditure data from the US to predict very high-cost patients in the top 5 cost percentiles, among the general population. Samples are derived from the Medical Expenditure Panel Survey’s Household Component data for 2006-2008 including 98,175 records. After pre-processing, partitioning and balancing the data, the final MEPS dataset with 31,704 records is modeled by Decision Trees (including C5.0 and CHAID), Neural Networks. Multiple predictive models are built and their performances are analyzed using various measures including correctness accuracy, G-mean, and Area under ROC Curve. We conclude that the CHAID tree returns the best G-mean and AUC measures for top performing predictive models ranging from 76% to 85%, and 0.812 to 0.942 units, respectively. Among a primary set of 66 attributes, the best predictors to estimate the top 5% high-cost population include individual’s overall health perception, history of blood cholesterol check, history of physical/sensory/mental limitations, age, and history of colonic prevention measures. It is worthy to note that we do not consider number of visits to care providers as a predictor since it has a high correlation with the expenditure, and does not offer a new insight to the data (i.e. it is a trivial predictor). We predict high-cost patients without knowing how many times the patient was visited by doctors or hospitalized. Consequently, the results from this study can be used by policy makers, health planners, and insurers to plan and improve delivery of health services.

In this research, we apply data mining techniques to a nationally-representative expenditure data from the US to predict very high-cost patients in the top 5 cost percentiles, among the general population. Samples are derived from the Medical Expenditure Panel Survey’s Household Component data for 2006-2008 including 98,175 records. After pre-processing, partitioning and balancing the data, the final MEPS dataset with 31,704 records is modeled by Decision Trees (including C5.0 and CHAID), Neural Networks. Multiple predictive models are built and their performances are analyzed using various measures including correctness accuracy, G-mean, and Area under ROC Curve. We conclude that the CHAID tree returns the best G-mean and AUC measures for top performing predictive models ranging from 76% to 85%, and 0.812 to 0.942 units, respectively. Among a primary set of 66 attributes, the best predictors to estimate the top 5% high-cost population include individual’s overall health perception, history of blood cholesterol check, history of physical/sensory/mental limitations, age, and history of colonic prevention measures. It is worthy to note that we do not consider number of visits to care providers as a predictor since it has a high correlation with the expenditure, and does not offer a new insight to the data (i.e. it is a trivial predictor). We predict high-cost patients without knowing how many times the patient was visited by doctors or hospitalized. Consequently, the results from this study can be used by policy makers, health planners, and insurers to plan and improve delivery of health services.

No description available.

No description available.

Intrinsic value prediction is a Natural Language Processing (NLP) problem consisting in determining a numerical value contained implicitly and non-trivially in a text. In this project, we introduce the SWORDSMAN model (Sentence and Word-level Oracle for Research Documents by Semantic Multi-scale ANalysis), a deep neural network architecture based on transformers whose goal is to predict the cost of research projects from the analysis of their abstract. SWORDSMAN is built on a hybrid structure based on two branches in order to conduct a multi-scale analysis by combining the strengths of global and local perspectives to extract more relevant information from these texts. The local branch uses Convolution Neural Networks (CNNs) to analyse abstracts at fine-grained word level and bring more nuance to the understanding of the context of occurrence of key terms, while the global branch combines Sentence Transformers and Radial Basis Functions (RBFs) to process these abstracts at a higher level to identify the overall context of the project, while being more focused on the content than the form of the data. The joint use of these models allows SWORDSMAN to have a better capacity to understand complex data by using this analysis at different levels of granularity to present a better estimation accuracy. / Förutsägelse av inneboende värde är ett problem inom Natural Language Processing (NLP) som består i att bestämma ett numeriskt värde som finns implicit och icke-trivialt i en text. I det här projektet introducerar vi SWORDSMAN-modellen (Sentence and Word-level Oracle for Research Documents by Semantic Multi-scale ANalysis), en djup neuronal nätverksarkitektur baserad på transformatorer vars mål är att förutsäga kostnaden för forskningsprojekt utifrån analysen av deras abstrakt. SWORDSMAN bygger på en hybridstruktur baserad på två grenar för att genomföra en analys i flera skalor genom att kombinera styrkorna hos globala och lokala perspektiv för att extrahera mer relevant information från dessa texter. I den lokala grenen används CNN-nätverk (Convolution Neural Networks) för att analysera sammanfattningar på finkornig ordnivå och ge mer nyans till förståelsen av sammanhanget för förekomsten av nyckeltermer, medan den globala grenen kombinerar meningstransformatorer och radiella basfunktioner (RBF) för att bearbeta dessa sammanfattningar på en högre nivå för att identifiera projektets övergripande sammanhang, samtidigt som den är mer inriktad på innehållet än på formen av uppgifterna. Den gemensamma användningen av dessa modeller gör det möjligt för SWORDSMAN att ha en bättre förmåga att förstå komplexa data genom att använda denna analys på olika granularitetsnivåer för att presentera en bättre skattningsnoggrannhet. / La prédiction de valeur intrinsèque est un problème de Traitement Automatique du Langage (TAL) consistant à déterminer une valeur numérique contenue de manière implicite et non triviale dans un texte. Dans ce projet, nous introduisons le modèle SWORDSMAN (Sentence and Word-level Oracle for Research Documents by Semantic Multi-scale ANalysis), une architecture de réseaux de neurones profonde basée sur les transformers dont le but est de prédire le coût de projets de recherche à partir de l’analyse de leur abstract. SWORDSMAN est bâti sur une structure hybride reposant sur deux branches afin de mener une analyse multi-échelles en combinant les forces de perspectives globale et locale pour extraire des informations plus pertinentes de ces textes. La branche locale utilise des réseaux de neurones de convolution (CNN) pour analyser les abstracts à l’échelle des mots et apporter plus de nuance à la compréhension du contexte d’apparition des termes clés, là où la branche globale combine Sentence Transformers et fonctions de base radiale (RBF) pour traiter ces abstracts à un plus haut niveau afin d’identifier le contexte général du projet, tout en étant plus focalisée sur le contenu que la forme des données. L’utilisation conjointe de ces modèles permet à SWORDSMAN de disposer d’une meilleure capacité de compréhension de données complexes en se servant de cette analyse à différents niveaux de granularité pour présenter une meilleure précision d’estimation.

With the rapid evolution in mobile telecommunications, there is a significant need for more accurate and efficient management of resources such as CPU, RAM, and bandwidth. This thesis utilizes customer usage data alongside machine learning algorithms to predict resource demands, enabling telecommunications service providers to optimize service quality and reduce unnecessary costs. This thesis investigates enhancing mobile network cost prediction and resource utilization by integrating customer behavior analysis using machine learning models. As a predictive model we employed various machine learning techniques, including Random Forest Regressor and Recurrent Neural Networks (LSTM and GRU), and can effectively predict resource needs based on user events. Among these models, the Random Forest Regressor performed the best. This model enhances operational efficiency by providing precise resource predictions within the dataset ranges. / Med den snabba utvecklingen inom mobiltelekommunikation finns det ett betydande behov av mer exakt och effektiv hantering av resurser som CPU, RAM och bandbred. Rapporten använder data om kundanvändning tillsammans med maskininlärningsalgoritmer för att förutsäga resursbehov, vilket möjliggör att telekommunikationsleverantörer kan optimera tjänstekvalitet och minska onödiga kostnader. Detta examensarbete undersöker hur förutsägelser av kostnader och resursanvändning i mobila nätverk kan förbättras genom att integrera analys av kundbeteende med maskininlärningsmodeller. Som en prediktiv modell använde vi olika maskininlärningstekniker, inklusive Random Forest Regressor och Recurrent Neural Networks (LSTM och GRU), effektivt kan förutsäga resursbehov baserat på användarhändelser. Bland dessa modeller presterade Random Forest Regressor bäst. Denna modell förbättrar den operativa effektiviteten genom att ge mer precisa resurs prediktion inom datamängdens intervaller.