Forecasting the Consumer Confidence Index with tree-based MIDAS regressions

The macroeconomic literature has recently uncovered the importance of the consumer confidence variations at driving business cycles. However, it remains a challenge to predict changes in agents'confidence by exploiting the information from ultra high-frequency sentiment data extracted from social media. Based on the mixed data sampling (MIDAS) literature, we propose a new MIDAS method that introduces regression tree-based algorithms into the MIDAS framework. Our method is more flexible at sampling high-frequency lagged regressors compared to existing MIDAS models with tightly parametrized functions of lags. In an out-of-sample forecasting exercise for the Consumer Confidence Index, our results reveal that (i) the proposed procedure exploits more fully the information from historical sentiment data and (ii) our method substantially improves the forecast accuracy and confirms the role of social media at affecting the consumer confidence.     

Automated data-driven prediction on aircraft Estimated Time of Arrival

4D trajectory prediction is the core element of the future air transportation system. It aims to improve the operational ability and the predictability of air traffic. In this paper, a novel automated data-driven framework to deal with the prediction of Estimated Time of Arrival (ETA) on the runway at the entry point of Terminal Manoeuvring Area (TMA) is introduced. The proposed framework mainly consists of data preprocessing and machine learning models. Firstly, the dataset is divided, analyzed, cleaned, and estimated. Then, the flights are clustered into partitions according to different runway-in-use (QFU). Several candidate machine learning models are trained and selected on the corresponding dataset of each QFU. Feature engineering is conducted to transform raw data into features. After that, the experiments are performed on real ADS-B data in Beijing TMA with nested cross validation. By comparing the prediction performance on the preprocessed and un-preprocessed datasets, the results demonstrate that the proposed data preprocessing is able to improve the data quality. It is also robust to outliers, missing data, and noise. Finally, an ensemble learning strategy named stacking is introduced. Compared to other individual models, the stacked model has a more complex structure and performs best in ETA prediction. This fact reveals that the framework proposed in this study could make accurate and reliable ETA predictions.     

Innovation analytics: Leveraging artificial intelligence in the innovation process

Artificial intelligence (AI) is about imbuing machines with a kind of intelligence that is mainly attributed to humans. Extant literature—coupled with our experiences as practitioners—suggests that while AI may not be ready to completely take over highly creative tasks within the innovation process, it shows promise as a significant support to innovation managers. In this article, we broadly refer to the derivation of computer-enabled, data-driven insights, models, and visualizations within the innovation process as innovation analytics. AI can play a key role in the innovation process by driving multiple aspects of innovation analytics. We present four different case studies of AI in action based on our previous work in the field. We highlight benefits and limitations of using AI in innovation and conclude with strategic implications and additional resources for innovation managers.     

Artificial intelligence and machine learning as business tools: A framework for diagnosing value destruction potential

Artificial intelligence (AI) and machine learning (ML) may save money and improve the efficiency of business processes, but these technologies can also destroy business value, sometimes with grave consequences. The inability to identify and manage that risk can lead some managers to delay the adoption of these technologies and thus prevent them from realizing their potential. This article proposes a new framework by which to map the components of an AI solution and to identify and manage the value-destruction potential of AI and ML for businesses. We show how the defining characteristics of AI and ML can threaten the integrity of the AI system’s inputs, processes, and outcomes. We then draw from the concepts of value-creation content and value-creation process to show how these risks may hinder value creation or even result in value destruction. Finally, we illustrate the application of our framework with an example of the deployment of an AI-powered chatbot in customer service, and we discuss how to remedy the problems that arise.     

Artificial intelligence: Building blocks and an innovation typology

The range of topics and the opinions expressed on artificial intelligence (AI) are so broad that clarity is needed on the the field’s central tenets, the opportunities AI presents, and the challenges it poses. To that end, we provide an overview of the six building blocks of artificial intelligence: structured data, unstructured data, preprocesses, main processes, a knowledge base, and value-added information outputs. We then develop a typology to serve as an analytic tool for managers grappling with AI’s influence on their industries. The typology considers the effects of AI-enabled innovations on two dimensions: the innovations’ boundaries and their effects on organizational competencies. The typology’s first dimension distinguishes between product-facing innovations, which influence a firm’s offerings, and process-facing innovations, which influence a firm’s operations. The typology’s second dimension describes innovations as either competence-enhancing or competence-destroying; the former enhances current knowledge and skills, whereas the latter renders existing skills and knowledge obsolete. This framework lets managers evaluate their markets, the opportunities within them, and the threats arising from them, providing valuable background and structure to important strategic decisions.     

Machines learn neuromarketing: Improving preference prediction from self-reports using multiple EEG measures and machine learning

A basic aim of marketing research is to predict consumers’ preferences and the success of marketing campaigns at the population-level. However, traditional marketing tools have various limitations, calling for novel measures to improve predictive power. In this study, we use multiple types of measures extracted from electroencephalography (EEG) recordings and machine learning (ML) algorithms to improve preference prediction based on self-reports alone. Subjects watched video commercials of six food products as we recorded their EEG activity, after which they responded to a questionnaire that served as a self-report benchmark measure. Thereafter, subjects made binary choices over the food products. We attempted to predict within-sample and population level preferences, based on subjects’ questionnaire responses and EEG measures extracted during the commercial viewings. We reached 68.5% accuracy in predicting between subjects’ most and least preferred products, improving accuracy by 4.07 percentage points compared to prediction based on self-reports alone. Additionally, EEG measures improved within-sample prediction of all six products by 20%, resulting in only a 1.91 root mean squared error (RMSE) compared to 2.39 RMSE with questionnaire-based prediction alone. Moreover, at the population level, assessed using YouTube metrics and an online questionnaire, EEG measures increased prediction by 12.7% and 12.6% respectively, compared to only a questionnaire-based prediction. We found that the most predictive EEG measures were frontal powers in the alpha band, hemispheric asymmetry in the beta band, and inter-subject correlation in delta and alpha bands. In summary, our novel approach, employing multiple types of EEG measures and ML models, offers marketing practitioners and researchers a valuable tool for predicting individual preferences and commercials’ success in the real world.     

A tale of two recommender systems: The moderating role of consumer expertise on artificial intelligence based product recommendations

Recommender systems are used in e-Commerce websites to make product recommendations or deliver personalized content to users. We constructed a beer recommendation program using review data from existing online community to test the hypotheses. This research aims to bridge the gap between marketing and computer science by investigating the moderating effects of consumer knowledge (expertise) on the performance and evaluation of two widely-used recommendation systems – user-based collaborative filtering and content-based. The results show that expert consumers prefer user-based collaborative filtering systems, whereas there is no difference between the two systems among novice consumers. Theoretical and managerial implications are discussed.     

Economic resilience in times of public health shock: The case of the US states

Does adopting social distancing policies amid a health crisis, e.g., COVID-19, hurt economies? Using a machine learning approach at the intermediate stage, we applied a generalized synthetic control method to answer this question. We utilize state policy response differences. Cross-validation, a machine learning approach, is used to produce the “counterfactual” for adopting states—how they “would have behaved” without lockdown orders. We categorize states with social distancing as the treatment group and those without as the control. We employ the state time-period for fixed effects, adjusting for selection bias and endogeneity. We find significant and intuitively explicable impacts on some states, such as West Virginia, but none at the aggregate level, suggesting that social distancing may not affect the entire economy. Our work implies a resilience index utilizing the magnitude and significance of the social distancing measures to rank the states' resilience. These findings help governments and businesses better prepare for shocks.     

Estimating entry counts and ATFM regulations during adverse weather conditions using machine learning

In recent years, convective weather has been the cause of significant delays in the European airspace. With climate experts anticipating the frequency and intensity of convective weather to increase in the future, it is necessary to find solutions that mitigate the impact of convective weather events on the airspace system. Analysis of historical air traffic and weather data will provide valuable insight on how to deal with disruptive convective events in the future. We propose a methodology for processing and integrating historic traffic and weather data to enable the use of machine learning algorithms to predict network performance during adverse weather. In this paper we develop regression and classification supervised learning algorithms to predict airspace performance characteristics such as entry count, number of flights impacted by weather regulations, and if a weather regulation is active. Examples using data from the Maastricht Upper Area Control Centre are presented with varying levels of predictive performance by the machine learning algorithms. Data sources include Demand Data Repository from EUROCONTROL and the Rapid Developing Thunderstorm product from EUMETSAT.     

A machine learning approach to cluster destination image on Instagram

Symbols are powerful in branding and marketing to represent tourist attractions. By bridging semiotics, marketing, and data science in the tourism context, this study uncovers the destination image based on Instagram photographs. This study constructed a novel methodological framework by evaluating different machine learning models to group textual information based on pictorial content. The results highlighted specific destination image clusters such as the wilderness and spirituality of alpine experiences. This information facilitates marketers' understanding of tourists’ preferences and movement. It also discloses blind spots that are less promoted by the marketers.