Evaluation Metrics and Transfer Learning for Sales Prediction

Sales prediction plays a huge role in modern business strategies. One of it's many use cases revolves around estimating the effects of promotions. While promotions generally have a positive effect on sales of the promoted product, they can also have a negative effect on those of other products. This phenomenon is calles sales cannibalisation. Sales cannibalisation can pose a big problem to sales forcasting algorithms. A lot of times, these algorithms focus on sales over time of a single product in a single store (a couple). This research focusses on using knowledge of a product across multiple different stores. To achieve this, we applied transfer learning on a neural model developed by Kantar Consulting to demo an approach to estimating the effect of cannibalisation. Our results show a performance increase of between 10 and 14 percent. This is a very good and desired result, and Kantar will use the approach when integrating this test method into their actual systems.

Application of active and transfer learning to wound image segmentation

Wound management is a fundamental task in standard clinical practice. Automated solutions already exist for humans, but there is a lack of applications on wound management for pets. The importance of a precise and efficient wound assessment is helpful to improve diagnosis and to increase the effectiveness of treatment plans for the chronic wounds. The goal of the research was to propose an automated pipeline capable of segmenting natural light-reflected wound images of animals. Two datasets composed by light-reflected images were used in this work: Deepskin dataset, 1564 human wound images obtained during routine dermatological exams, with 145 manual annotated images; Petwound dataset, a set of 290 wound photos of dogs and cats with 0 annotated images. Two implementations of U-Net Convolutioal Neural Network model were proposed for the automated segmentation. Active Semi-Supervised Learning techniques were applied for human-wound images to perform segmentation from 10% of annotated images. Then the same models were trained, via Transfer Learning, adopting an Active Semi- upervised Learning to unlabelled animal-wound images. The combination of the two training strategies proved their effectiveness in generating large amounts of annotated samples (94% of Deepskin, 80% of PetWound) with the minimal human intervention. The correctness of automated segmentation were evaluated by clinical experts at each round of training thus we can assert that the results obtained in this thesis stands as a reliable solution to perform a correct wound image segmentation. The use of Transfer Learning and Active Semi-Supervied Learning allows to minimize labelling effort from clinicians, even requiring no starting manual annotation at all. Moreover the performances of the model with limited number of parameters suggest the implementation of smartphone-based application to this topic, helping the future standardization of light-reflected images as acknowledge medical images.

Assessment of marine litter in the Western Adriatic Sea and its impact on marine biota

Due to its environmental, safety, health and socio-economic impacts, marine litter has been recognized as a 21st century global challenge, so that it has been included in Descriptor 10 of the EU MSFD. For its morphological features and anthropogenic pressures, the Adriatic Sea is very sensitive to the accumulation of debris, but data are inconsistent and fragmented. This thesis, in the framework of DeFishGear project, intents to assess marine litter on beaches and on seafloor in the Western Adriatic sea, and test if debris ingestion by fish occurs. Three beaches were sampled during two surveys in 2015. Benthic litter monitoring was carried out in the FAO GSA17 during fall 2014, using a rapido trawl. Litter ingestion was investigated through gut contents analysis of 260 fish belonging to 8 commercial species collected in Western Gulf of Venice. Average litter density on beaches was 1.5 items/m2 during spring, and decreased to 0.8 items/m2 in summer. Litter composition was heterogeneous and varied among sites, even if no significant differences were found. Most of debris consisted of plastic sheets, fragments, polystyrene pieces, mussels nets and cottons bud sticks, showing that sources are many and include aquaculture, land-based activities and local users of beaches. Average density of benthic litter was 913 items/Km2 (82 Kg/Km2). Plastic dominated in terms of numbers and weight, and consisted mainly of bags, sheets and mussel nets. The highest density was found close to the coast, and sources driving the major differences in litter distribution were mussel farms and shipping lanes. Litter ingestion occurred in 47% of examined fish, mainly consisting of fibers. Among species, S. pilchardus swallowed almost all debris categories. Findinds may provide a baseline to set the necessary measures to manage and minimize marine litter in the Western Adriatic region and to protect aquatic life from plastic pollution, even accounting the possible implications on human health.

Correlation between urban form and PM2.5 concentration — Evidence from the New York metropolitan Area and Shanghai city

With the development of the economy and society, air pollution has posed a huge threat to public health around the world, especially to people who live in urban areas. Typically, urban development patterns can be roughly divided into compact cities and urban sprawl. In recent years, the relationship between urban form and air quality (especially PM2.5) is gaining more and more attention from urban planners, environmentalists, and governments. This study is focusing on The New York metropolitan area and Shanghai city, which are both megacities but with different urban spatial forms. For both study areas,there are five main variables to measure the urban form metrics, naming Population Density, Artificial Land Area Per Ten Thousand People, Road Density, Green Land Area Ratio and Artificial Land Area Ratio. In addition, considering the impact of economic activities and public transportation, GDP per capita, Number of bus stop and Number of subway station are used as control variables. Based on the results of regression, a megacity like the New York metropolitan area with urban sprawl shows a low spatial correlation on PM2.5 concentration. Meanwhile, almost all the spatial form indicators effect on PM2.5 concentration is not significant. However, a compact megacity like Shanghai shows a diametrically opposite result. Urban form, especially population density, has a strong relationship with PM2.5 concentration. It can be predicted that a reduction in population density would lead to significant improvements on decrease the PM2.5 concentration in Shanghai. Meanwhile, increasing the ratio of green land and construction area per capita will get a positive influence on reducing PM2.5 concentration as well. Road density is not a significant factor for a megacity in both two urban forms. The way and type of energy used by vehicles on megacities maybe more critical.