Industrial Engineering & Management

About 3 years ago, my boss decided that it’s time to leverage the superpowers of data. So, I was the first data scientist, a unicorn, amongst 6600 colleges at Aurubis. The primary task was to introduce, to explain, promote and establish data science skillset within the organization. Old industry, like metallurgy and mining, are not the typical examples of successful digital transformation because the related business models are extremely stable, even in the era of hyper-innovation. At least this is what some people believe, and it’s partly true, because for some branches, there is no burning platform for digitization, and hence, the change process is inert. Data science is the fundamental component of digital transformation. Our contribution to the change has a huge impact because we can extract the value from the data and generate the business value, to show people what can be done when the data is there and valid.

I learned that most valuable, essential skills to succeed in  our business are not necessarily programming and statistics. We all have training on data science methods at its best. The two must have skills are resilience and communication. Whenever you start something new, you will fail. You must be and stay resilient to rise strongly. Moreover, in the business world is the ability to communicate - tell data-based stories, to visualize and to promote them is crucial. As a data scientist you can only be as good as your communications skills are, since you need to persuade others to make decisions or help to build products based on your analyses. Finally, dare to start simple. When you introduce data science in the industry, you start on the brown field. Simple use cases and projects like metrics, dashboards, reports, historical analysis help you to understand the business model and to assess where is your contribution to success of the company. This is the key to data science success, not only in the multimetal but everywhere else as well.

Commonly known by the term “big data”, Data Science is the study of the generalizable extraction of knowledge from data. It assesses the behaviour of data in a controlled, logic-led, responsive environment for deriving automated solutions and prognostic models for a given situation, problem or business objective. From Tinder to Facebook; LinkedIn to various online giants like Amazon and Google, Data science is playing a pivotal role in making the data scientist the new sought-after job in the market. Using large amounts of data for decision making has become practical now, with industries hiring qualified data scientists to handle a wide range of unprocessed data to come up with modern workable solutions catering to their respective market. Gone are the days when companies used to work on software like Excel only to analyse and store data. Even at that time, only some intelligent ventures worked with SPSS and strata

The invasive Parthenium weed (Parthenium hyterophorus) adversely affects animal and human health, agricultural productivity, rural livelihoods, local and national economies, and the environment. Its fast spreading capability requires consistent monitoring for  adoption of relevant mitigation approaches, potentially through remote sensing. To date, studies that have endeavoured to map the Parthenium weed have commonly used popular classification algorithms that include Support vector machines and Random forest classifiers, which do not capture the complex structural characteristics of the weed. Furthermore, determination of site or data specific algorithms, often achieved through intensive comparison of algorithms, is often laborious and time consuming. Also, selected algorithms may not be optimal on datasets collected in other sites. Hence,  this study adopted the Tree-based Pipeline Optimization Tool (TPOT), an automated machine learning   approach that can be used to overcome high data variability during the classification process. Using Sentinel-2 and Landsat 8 imagery to map Parthenium weed, wee compared the outcome of the TPOT to the best performing and optimized algorithm selected from sixteen classifiers on different training datasets.  Results showed that the TPOT model yielded a higher overall classification accuracy (88.15%) using Sentinel-2 and 74 % using Landsat 8, accuracies that were higher than the commonly used robust classifiers. This study is the first to  demonstrate the value of TPOT in mapping Parthenium weed infestations using satellite imagery. Its adoption would therefore be useful in limiting human intervention while optimising classification accuracies for mapping invasive plants. Based on these findings, we propose TPOT as an efficient method for selecting and tuning algorithms for Parthenium discrimination and monitoring, and indeed general vegetation mapping.

Tree-based Pipeline Optimization Tool (TPOT) (Olson and Moore 2016) is a novel AutoML that applies genetic programming (GP) to optimize machine learning pipelines of the sklearn python library for classification and regression problems. The following pipeline operators are implemented in TPOT: pre-processors, decomposition, feature selection, and models. During the optimization process, subsets of the ML pipelines are defined as GP primitives, which are organized as in a tree structure to form individuals. To obtain the optimal combination of processes, GP optimizes the number and order of pipeline operators, as well as each operator’s parameters (Sohn, Olson, and Moore 2017). More details on the tool can be found in (Olson and Moore 2016). An example of TPOT workflow is illustrated in Figure 2. In this study, the choice of TPOT parameters was premised on the assumption that better results are achieved with more central processing unit (CPU) time allocated (Hutter, Kotthoff, and Vanschoren 2019). Therefore, parameters such as ‘generations’, ‘population_size’ and ‘verbosity’ were set to 500, 100 and 2 respectively. Furthermore a ‘random_state’ parameter was added to the python code containing the best pipeline generated by TPOT to allow replication     

The invasive Parthenium weed (Parthenium hyterophorus) adversely affects animal and human health, agricultural productivity, rural livelihoods, local and national economies, and the environment. Its fast spreading capability requires consistent monitoring for  adoption of relevant mitigation approaches, potentially through remote sensing. To date, studies that have endeavoured to map the Parthenium weed have commonly used popular classification algorithms that include Support vector machines and Random forest classifiers, which do not capture the complex structural characteristics of the weed. Furthermore, determination of site or data specific algorithms, often achieved through intensive comparison of algorithms, is often laborious and time consuming. Also, selected algorithms may not be optimal on datasets collected in other sites. Hence,  this study adopted the Tree-based Pipeline Optimization Tool (TPOT), an automated machine learning   approach that can be used to overcome high data variability during the classification process. Using Sentinel-2 and Landsat 8 imagery to map Parthenium weed, wee compared the outcome of the TPOT to the best performing and optimized algorithm selected from sixteen classifiers on different training datasets.  Results showed that the TPOT model yielded a higher overall classification accuracy (88.15%) using Sentinel-2 and 74 % using Landsat 8, accuracies that were higher than the commonly used robust classifiers. This study is the first to  demonstrate the value of TPOT in mapping Parthenium weed infestations using satellite imagery. Its adoption would therefore be useful in limiting human intervention while optimising classification accuracies for mapping invasive plants. Based on these findings, we propose TPOT as an efficient method for selecting and tuning algorithms for Parthenium discrimination and monitoring, and indeed general vegetation mapping.

Tree-based Pipeline Optimization Tool (TPOT) (Olson and Moore 2016) is a novel AutoML that applies genetic programming (GP) to optimize machine learning pipelines of the sklearn python library for classification and regression problems. The following pipeline operators are implemented in TPOT: pre-processors, decomposition, feature selection, and models. During the optimization process, subsets of the ML pipelines are defined as GP primitives, which are organized as in a tree structure to form individuals. To obtain the optimal combination of processes, GP optimizes the number and order of pipeline operators, as well as each operator’s parameters (Sohn, Olson, and Moore 2017). More details on the tool can be found in (Olson and Moore 2016). An example of TPOT workflow is illustrated in Figure 2. In this study, the choice of TPOT parameters was premised on the assumption that better results are achieved with more central processing unit (CPU) time allocated (Hutter, Kotthoff, and Vanschoren 2019). Therefore, parameters such as ‘generations’, ‘population_size’ and ‘verbosity’ were set to 500, 100 and 2 respectively. Furthermore a ‘random_state’ parameter was added to the python code containing the best pipeline generated by TPOT to allow replication     

Forecasting in time-series data is at the core of various business decision making activities. One key characteristic of many practical time series data of different business metrics such as orders, revenue, is the presence of irregular yet moderately frequent spikes of very high intensity, called extreme observation. Forecasting such spikes accurately is crucial for various business activities such as workforce planning, financial planning, inventory planning. Traditional time series forecasting methods such as ARIMA, BSTS, are not very accurate in forecasting extreme spikes. Deep Learning techniques such as variants of LSTM tend to perform only marginally better than these traditional techniques. The underlying assumption of thin tail of data distribution is one of the primary reasons for such models to falter on forecasting extreme spikes as moderately frequent extreme spikes result in heavy tail of the distribution. On the other hand, literatures, proposing methods to forecast extreme events in time series, focused mostly on extreme events but ignored overall forecasting accuracy. We attempted to address both these problems by proposing a technique where we considered a time series signal with extreme spikes as the superposition of two independent signals - (1) a stationary time series signal without extreme spike (2) a shock signal consisting of near-zero values most of the time along with few spikes of high intensity. We modelled the above two signals independently to forecast values for the original time series signal. Experimental results show that the proposed technique outperforms existing techniques in forecasting both normal and extreme events.  A tempest flood hindcast for the west shoreline of Canada was produced for the time frame 1980–2016 utilizing a 2D nonlinear barotropic Princeton Ocean Model constrained by hourly Climate Forecast System Reanalysis wind and ocean level pressing factor. Approval of the displayed storm floods utilizing tide measure records has shown that there are broad zones of the British Columbia coast where the model doesn't catch the cycles that decide the ocean level fluctuation on intraseasonal and interannual time scales. A portion of the inconsistencies are connected to enormous scope variances, for example, those emerging from significant El Niño and La Niña occasions. By applying a change in accordance with the hindcast utilizing a sea reanalysis item that consolidates enormous scope ocean level changeability and steric impacts, the difference of the mistake of the changed floods is fundamentally decreased (by up to half) contrasted with that of floods from the barotropic model. The significance of baroclinic elements and steric impacts to exact tempest flood anticipating in this waterfront area is illustrated, just like the need to consolidate decadal-scale, bowl explicit maritime inconstancy into the assessment of outrageous beach front ocean levels. The outcomes improve long haul extraordinary water level gauges and stipends for the west shoreline of Canada without long haul tide measure records information.