Browsing by Subject "Machine learning"

Now showing 1 - 12 of 12

Bi-objective optimization of nutrient intake and performance of broiler chickens using Gaussian process regression and genetic algorithm
(2023) Ahmadi, Hamed; Rodehutscord, Markus; Siegert, Wolfgang
This study investigated whether quantifying the trade-off between the maxima of two response traits increases the accuracy of diet formulation. To achieve this, average daily weight gain (ADG) and gain:feed ratio (G:F) responses of 7–21-day-old broiler chickens to the dietary supply of three nutrients (intake of digestible glycine equivalents, digestible threonine, and total choline) were modeled using a newly developed hybrid machine learning-based method of Gaussian process regression and genetic algorithm. The dataset comprised 90 data lines. Model-fit-criteria indicated a high model adjustment and no prediction bias of the models. The bi-objective optimization scenarios through Pareto front revealed the trade-off between maximized ADG and maximized G:F and provided information on the needed input of the three nutrients that interact with each other to achieve the trade-off scenarios. The trade-off scenarios followed a nonlinear pattern. This indicated that choosing target values intermediate to maximized ADG and G:F after single-objective optimization is less accurate than feed formulation after quantifying the trade-off. In conclusion, knowledge of the trade-off between maximized ADG and maximized G:F and the needed nutrient inputs will help feed formulators to optimize their feed with a more holistic approach.
Consumer prices : effects of learning algorithms and pandemic-related policy measures
(2023) Buchali, Katrin; Schwalbe, Ulrich
When it comes to product prices, two major topics have dominated the public debate in recent years: One is pricing with the help of artificial intelligence, and the other is the price level, which has risen more than usual with the onset of the COVID-19 pandemic. Higher prices create a loss of consumer surplus and possibly total welfare, which is the reason this topic has become ubiquitous in political discussions. This dissertation contributes to the debate by extending the existing literature on algorithmic pricing, which is said to facilitate personalized pricing, as well as collusive behavior and to enhance the general understanding of how government measures enforced during the COVID-19 pandemic contributed to (short-time) price developments. Thereby, the first part of the thesis addresses the concern that tacit collusion might occur if firms employ learning algorithms, as several simulation studies have demonstrated that algorithms using reinforcement learning are able to coordinate their pricing behavior and, as a result, achieve a collusive outcome without having been programmed for it. We discuss several conceptual challenges as well as challenges in the real-world application of algorithms and show by or own simulations that resulting market prices strongly depend on the type of algorithm or heuristic that is used by the firms to set prices. In the subsequent part of the thesis we examine how a self-learning pricing algorithm performs when faced with inequity-averse consumers. From our simulations we can conclude that consumers sense of fairness, which have prevented firms from engaging in price discrimination in the past years, can be incorporated into firms pricing decisions with the help of learning algorithms, making differential pricing strategies more feasible. The discussion surrounding the above-average price levels in many countries during the COVID-19 pandemic is extended in the third part of the thesis. We present empirical evidence for the impact of government-imposed restrictions and, as a consequence of their enforcement, reduced mobility on consumer prices during the COVID-19 pandemic. We show that the stringency of government measures had a positive and significant impact on consumer prices mainly in the food sector, which means that more stringent measures induced higher consumer prices in these categories.
CortexVR: Immersive analysis and training of cognitive executive functions of soccer players using virtual reality and machine learning
(2022) Krupitzer, Christian; Naber, Jens; Stauffert, Jan-Philipp; Mayer, Jan; Spielmann, Jan; Ehmann, Paul; Boci, Noel; Bürkle, Maurice; Ho, André; Komorek, Clemens; Heinickel, Felix; Kounev, Samuel; Becker, Christian; Latoschik, Marc Erich
Goal: This paper presents an immersive Virtual Reality (VR) system to analyze and train Executive Functions (EFs) of soccer players. EFs are important cognitive functions for athletes. They are a relevant quality that distinguishes amateurs from professionals. Method: The system is based on immersive technology, hence, the user interacts naturally and experiences a training session in a virtual world. The proposed system has a modular design supporting the extension of various so-called game modes. Game modes combine selected game mechanics with specific simulation content to target particular training aspects. The system architecture decouples selection/parameterization and analysis of training sessions via a coaching app from an Unity3D-based VR simulation core. Monitoring of user performance and progress is recorded by a database that sends the necessary feedback to the coaching app for analysis. Results: The system is tested for VR-critical performance criteria to reveal the usefulness of a new interaction paradigm in the cognitive training and analysis of EFs. Subjective ratings for overall usability show that the design as VR application enhances the user experience compared to a traditional desktop app; whereas the new, unfamiliar interaction paradigm does not negatively impact the effort for using the application. Conclusion: The system can provide immersive training of EF in a fully virtual environment, eliminating potential distraction. It further provides an easy-to-use analyzes tool to compare user but also an automatic, adaptive training mode.
Electronic nose for the rapid detection of deoxynivalenol in wheat using classification and regression trees
(2022) Camardo Leggieri, Marco; Mazzoni, Marco; Bertuzzi, Terenzio; Moschini, Maurizio; Prandini, Aldo; Battilani, Paola
Mycotoxin represents a significant concern for the safety of food and feed products, and wheat represents one of the most susceptible crops. To manage this issue, fast, reliable, and low-cost test methods are needed for regulated mycotoxins. This study aimed to assess the potential use of the electronic nose for the early identification of wheat samples contaminated with deoxynivalenol (DON) above a fixed threshold. A total of 214 wheat samples were collected from commercial fields in northern Italy during the periods 2014–2015 and 2017–2018 and analyzed for DON contamination with a conventional method (GC-MS) and using a portable e-nose “AIR PEN 3” (Airsense Analytics GmbH, Schwerin, Germany), equipped with 10 metal oxide sensors for different categories of volatile substances. The Machine Learning approach “Classification and regression trees” (CART) was used to categorize samples according to four DON contamination thresholds (1750, 1250, 750, and 500 μg/kg). Overall, this process yielded an accuracy of >83% (correct prediction of DON levels in wheat samples). These findings suggest that the e-nose combined with CART can be an effective quick method to distinguish between compliant and DON-contaminated wheat lots. Further validation including more samples above the legal limits is desirable before concluding the validity of the method.
Empirical essays on initial public offerings
(2022) Reiff, Annika; Tykvová, Tereza
This dissertation builds on and extends previous IPO literature by analyzing unresolved questions with regard to the phenomenon of IPO withdrawals and the effect of IPOs on industry rivals. After a short introductory chapter, chapter 2 contributes to the analysis of IPO withdrawal by taking a data-driven and forward-looking perspective. In particular, it applies two machine learning methods, namely lasso and random forest, to predict IPO withdrawal and compares the performance of both models to the performance of a logistic regression model. Results show that random forest predicts IPO withdrawal quite well and outperforms lasso and logit with regard to in-sample prediction and cross-sectional out-of-sample prediction. However, all models fail substantially when trying to predict future IPO withdrawal. One explanation for this puzzling finding is the presence of concept drift – a change in the relationship between the predictors and IPO withdrawal over time. Further, the study contributes to the clarification of the question of which variables are most important to predict IPO withdrawal by exploiting certain features of the machine learning methods and considering a vast selection of different predictors. Market characteristics at filing seem to be the most important variables for prediction in all models, while corporate governance and intermediary characteristics seem to be less important. Closely related to the second chapter, the third chapter takes a more theory-based perspective on IPO withdrawal. This chapter is co-authored with Tereza Tykvová and a reviewed version is published in the Journal of Corporate Finance. It addresses the question whether certain factors, particularly high-quality corporate governance and VC backing, may serve as signals for investors and can thus reduce the withdrawal probability, especially in risky market environments. The latter argument is based on the assumption that investors are especially careful in these situations and thus signals might be especially meaningful. Results from an interaction-term analysis suggest that corporate governance characteristics, like large and experienced boards, are indeed able to reduce the withdrawal probability in highly volatile markets. However, this finding does not hold true for VC backing per se. We therefore delve deeper into the effect of VCs by distinguishing three VC characteristics: syndicated vs. stand-alone VCs, domestic vs. foreign VCs, and VCs with high vs. low reputation. The analysis reveals that local VCs and VC syndication tend to reduce the withdrawal probability, particularly in highly volatile markets, which supports the signaling explanation. In contrast, the withdrawal probability for firms backed by reputable VCs tend to be lower only in less volatile and not in highly volatile markets. One explanation for this finding could be that these firms rather follow a dual-track strategy or postpone the IPO more likely in highly volatile markets than in less volatile markets. Chapter 4 moves away from IPO withdrawals towards the consideration of intra-industry effects of IPOs. Irrespective of the question of whether to withdraw or complete an IPO after filing, an IPO filing might influence its industry rivals. In order to analyze the mechanisms behind the effects of IPO filings on industry rivals more closely, I apply a new two-step-methodology which consists of an event study in the first step and a Difference-in-Difference analysis in the second step. This methodology allows to separately test for the existence of a competition and an information effect. The rationale of the competition effect is that by going public, firms gain some kind of competitive advantage over their industry rivals thereby increasing the competitive pressure in the industry and harm their rivals. The idea behind the information effect is that an IPO filing does not only deliver information about the IPO firm but about also about the industry in which it operates. In this connection, the information effect could either induce positive (by signaling good growth prospects) or negative (by foreshadowing future negative industry trends or revealing that the industry is overvalued) valuation effects on industry rivals. Results provide evidence for the existence of the competition effect, suggesting that IPO filings tend to harm industry rivals to a certain extent. In contrast, results do not provide sufficient evidence for the existence of the information effect. However, the lack of evidence for an aggregate information effect could also be the result of a positive effect on some but a negative effect on other rival firms which cancel each other out. Finally, chapter 5 concludes with a summary and provides and outlook for future research in the field of IPOs.
Gut microbiota patterns predicting long-term weight loss success in individuals with obesity undergoing nonsurgical therapy
(2022) Bischoff, Stephan C.; Nguyen, Nguyen K.; Seethaler, Benjamin; Beisner, Julia; Kügler, Philipp; Stefan, Thorsten
The long-term success of nonsurgical weight reduction programs is variable; thus, predictors of outcome are of major interest. We hypothesized that the intestinal microbiota known to be linked with diet and obesity contain such predictive elements. Methods: Metagenome analysis by shotgun sequencing of stool DNA was performed in a cohort of 15 adults with obesity (mean body mass index 43.1 kg/m2) who underwent a one-year multidisciplinary weight loss program and another year of follow-up. Eight individuals were persistently successful (mean relative weight loss 18.2%), and seven individuals were not successful (0.2%). The relationship between relative abundancies of bacterial genera/species and changes in relative weight loss or body mass index was studied using three different statistical modeling methods. Results: When combining the predictor variables selected by the applied statistical modeling, we identified seven bacterial genera and eight bacterial species as candidates for predicting success of weight loss. By classification of relative weight-loss predictions for each patient using 2–5 term models, 13 or 14 out of 15 individuals were predicted correctly. Conclusions: Our data strongly suggest that gut microbiota patterns allow individual prediction of long-term weight loss success. Prediction accuracy seems to be high but needs confirmation by larger prospective trials.
Online monitoring of sourdough fermentation using a gas sensor array with multivariate data analysis
(2023) Anker, Marvin; Yousefi-Darani, Abdolrahim; Zettel, Viktoria; Paquet-Durand, Olivier; Hitzmann, Bernd; Krupitzer, Christian
Sourdough can improve bakery products’ shelf life, sensory properties, and nutrient composition. To ensure high-quality sourdough, the fermentation has to be monitored. The characteristic process variables for sourdough fermentation are pH and the degree of acidity measured as total titratable acidity (TTA). The time- and cost-intensive offline measurement of process variables can be improved by utilizing online gas measurements in prediction models. Therefore, a gas sensor array (GSA) system was used to monitor the fermentation process of sourdough online by correlation of exhaust gas data with offline measurement values of the process variables. Three methods were tested to utilize the extracted features from GSA to create the models. The most robust prediction models were achieved using a PCA (Principal Component Analysis) on all features and combined two fermentations. The calibrations with the extracted features had a percentage root mean square error (RMSE) from 1.4% to 12% for the pH and from 2.7% to 9.3% for the TTA. The coefficient of determination (R2) for these calibrations was 0.94 to 0.998 for the pH and 0.947 to 0.994 for the TTA. The obtained results indicate that the online measurement of exhaust gas from sourdough fermentations with gas sensor arrays can be a cheap and efficient application to predict pH and TTA.
Screen for collusive behavior : a machine learning approach
(2024) Bantle, Melissa
The paper uses a machine learning technique to build up a screen for collusive behavior. Such tools can be applied by competition authorities but also by companies to screen the behavior of their suppliers. The method is applied to the German retail gasoline market to detect anomalous behavior in the price setting of the filling stations. Therefore, the algorithm identifies anomalies in the data-generating process. The results show that various anomalies can be detected with this method. These anomalies in the price setting behavior are then discussed with respect to their implications for the competitiveness of the market.
Sentiment analysis in electronic negotiations
(2017) Körner, Michael; Schoop, Mareike
The thesis analyzes the applicability of methods of Sentiment Analysis and Predictive Analytics on textual communication in electronic negotiation transcripts. In particular, the thesis focuses on examining whether an automatic classifier can predict the outcome of ongoing, asynchronous electronic negotiations with sufficient accuracy. When combined with influencing factors leading to the specific classification decision, such a classification model could be incorporated into a Negotiation Support System in order to proactively intervene in ongoing negotiations it judges as likely to fail and then to give advice to the negotiators to prevent negotiation failure. To achieve this goal, an existing data set of electronic negotiations was used in a first study to create a Sentiment Lexicon, which tracks verbal indicators for utterances of positive and, respectively, negative polarity. This lexicon was subsequently combined with a simplified, feature-based representation of electronic negotiation transcripts which was then used as training data for various machine learning classifiers in order to let them determine the outcome of the negotiations based on the transcripts in a second study. Here, complete negotiation transcripts were classified as well as partial transcrips in order to assess classification quality in ongoing negotiations. The third study of the thesis sought to refine the classification model with respect to sentence-based granularity. To this end, human coders were classifying negotiation sentences regarding their subjectivity and polarity. The results of this content analysis approach were then used to train sentence-level subjectivity and polarity classifiers. The fourth and final study analyzed different aggregation methods for these sentence-level classification results in order to support the classifiers on negotiation granularity. Different aggregation and classification models were discussed, applied to the negotiation data and subsequently evaluated. The results of the studies show that it is possible to a certain degree to use a sentiment-based representation of negotiation data to automatically determine negotiation outcomes. In combination with the sentence-based classification models, negotiation classification quality increased further. However, this improvement was only found to be significant for complete negotiation transcripts. If only partial transcripts are used – specifically to simulate an ongoing negotiation scenario – the models tend to behave more erratic and classifcation quality depletes. This result yields the assumption that polarized utterances (positive as well as negative) only carry unequivocal information (with respect to the outcome) towards the end of the negotiation. During the negotiation, the influence of these utterances becomes more ambiguous, hence decreasing classification accuracy on models using a representation based on sentiments. Regarding the original goal of the thesis, which is to provide a basic means to support ongoing negotiations, this means that supporting mechanisms employed by a Negotiation Support System should focus on moderation techniques and resolving of potentially conflicting situations. Approaches that could be used to employ further conflict diagnosis in interaction with the negotiators are given in the final chapter of the thesis, as well as a discussion of potential recommendations and advice the system could give and lastly, approaches to visualize the classification data to the negotiators.
The implications of automation for economic growth and the labor share
(2016) Prettner, Klaus
We introduce automation into a standard model of capital accumulation and show that (i) there is the possibility of perpetual growth, even in the absence of technological progress; (ii) the long-run economic growth rate declines with population growth, which is consistent with the available empirical evidence; (iii)there is a unique share of savings diverted to automation that maximizes long-run growth; (iv) the labor share declines with automation to an extent that fits to the observed pattern over the last decades.
UAV remote sensing for high-throughput phenotyping and for yield prediction of Miscanthus by machine learning techniques
(2022) Impollonia, Giorgio; Croci, Michele; Ferrarini, Andrea; Brook, Jason; Martani, Enrico; Blandinières, Henri; Marcone, Andrea; Awty-Carroll, Danny; Ashman, Chris; Kam, Jason; Kiesel, Andreas; Trindade, Luisa M.; Boschetti, Mirco; Clifton-Brown, John; Amaducci, Stefano
Miscanthus holds a great potential in the frame of the bioeconomy, and yield prediction can help improve Miscanthus’ logistic supply chain. Breeding programs in several countries are attempting to produce high-yielding Miscanthus hybrids better adapted to different climates and end-uses. Multispectral images acquired from unmanned aerial vehicles (UAVs) in Italy and in the UK in 2021 and 2022 were used to investigate the feasibility of high-throughput phenotyping (HTP) of novel Miscanthus hybrids for yield prediction and crop traits estimation. An intercalibration procedure was performed using simulated data from the PROSAIL model to link vegetation indices (VIs) derived from two different multispectral sensors. The random forest algorithm estimated with good accuracy yield traits (light interception, plant height, green leaf biomass, and standing biomass) using 15 VIs time series, and predicted yield using peak descriptors derived from these VIs time series with root mean square error of 2.3 Mg DM ha−1. The study demonstrates the potential of UAVs’ multispectral images in HTP applications and in yield prediction, providing important information needed to increase sustainable biomass production.
Using machine learning for supply and demand predictions in the German milk market
(2023) Baaken, Dominik; Hess, Sebastian
The German milk market is driven by various ongoing trends on both the domestic supply and the international demand side. This results in increasingly volatile prices, as well as increasing production costs, and both risks continue to induce dairy farms going out of business. Therefore market participants have expressed a desire for reliable forecasting tools at the regional level in order to be able to make strategic and operational decisions with greater planning certainty. However, such forecasting models at the farm or regional level do not currently exist or are not publicly available. This dissertation fills this research gap by developing a forecasting model for predicting regional milk production in Lower Saxony. The first of four research chapters, Chapter 3, compares five different Machine Learning (ML) models and a traditional linear regression (OLS) model based on time trends, direct and indirect weather influences, and price events. The ML models show advantages in forecast accuracy, in particular ML methods outperform econometric modelling in predicting non-linear developments induced by investment. Furthermore, differences in the efficiency of the methods are apparent: while comparable estimation approaches achieve similar accuracies, the training speed of the models varies considerably. Chapter 4 presents the relationship between seasonal weather conditions and seasonal milk production. This chapter incorporates the influences of direct and indirect weather conditions as well as time and price trends into the model. A Fixed Effects (FE) estimator is used to model quarterly milk production for a panel dataset from Lower Saxony. The results mainly illustrate the influence of farm decisions on milk production, which is stronger than the influence of weather conditions. Contrary to expectations, the influence of weather conditions during the growing season cannot be significantly demonstrated. Instead, there is a positive effect of warmer and drier weather in almost all quarters except autumn. Chapters 5 and 6 shift the focus to the demand side of the German milk market, examining in particular the sale of raw milk from vending machines. As farmers seek alternative sales channels, on-farm vending machines offer an opportunity for additional income. Chapter 5 develops a forecasting model based on a nationwide survey and the Xtreme Gradient Boosting (XGB) algorithm. The model achieves sufficiently accurate values to qualify as a practical tool, allowing indecisive farm managers to input their own values into the model and thus secure their investment decision. The influence of the variables on the prediction is investigated using SHapley Additive exPlanation (SHAP) values, indicating that sales of raw milk from vending machines are influenced less by individual marketing measures than by various location factors such as population density, proximity to a city, and location along a road with commuter traffic. It can be concluded that there is additional sales potential if farmers would be allowed to place the vending machine in an optimal location away from the farm. Chapter 6 analyses consumer behaviour through a survey in Germany, using seemingly unrelated regression (SUR) to model willingness to pay (WTP) and frequency of purchase. The results suggest that in this form of marketing, consumers especially value a ‘fair’ price for the producer and are less price-sensitive. On average, customers’ WTP is higher than the current milk price and varies between consumer groups. Consumers with a closer connection to milk production are willing to pay more for raw milk but purchase it less frequently. It also appears that as consumers get older, they are more likely to buy raw milk but are less willing to pay for it. Tailoring marketing activities based on consumer characteristics can increase the efficiency of additional sales channels. Overall, this dissertation demonstrates the potential applications and limitations of ML methods for considering supply and demand in the German milk market. The forecasting models can serve as a potential tool for farmers to better weight strategic and operational decisions, thus contributing to more efficient agriculture.