This report is about big data concepts and the technologies used in big data. There are three tasks presented in this report with big data being the central aspect of all discussions.

The first part of the report is on data mining aspect of big data. Predictive analysis is performed on a large weather data set to predict if there will be rain tomorrow. The analysis is performed using RapidMiner software tool. The outcome of the use of RapidMiner is to apply statistical techniques, predictive modelling and machine learning techniques on the dataset in the prediction of rainfall. This kind of predictive analysis about the weather condition is helpful for weather depended companies like construction company so that they can manage their inventory, meet their labour needs, handle their supply chain, keep their people safe and delivery their promises to their clients.

The second task of the report discusses about the components of a proposed big data architecture for a government owned water utility company. The architecture covers the big data technologies used to handle data having characteristics like big volume, greater velocity, exhibiting large data variability.  The proposed architecture provides the necessary IT infrastructure to help the company in its overall operations with its revenue loss due to incorrect billing, theft, outage and demand predictions and better customer service. The report also covers literature surrounding the privacy, security and ethical concerns of the big data usage.

The last task of the report is use of visualization tool Tableau to create reports on Aviation Wildlife Strike. Interactive report is created to answer questions about the about hidden facts from the available data. The report gives insights about the cost of the wild life strikes, the strikes distributed during the time of day, damage caused to aircraft and to humans. The information is helpful for the airlines to understand the shortcomings of the existing systems and improve their operation efficiency.

The report in its concise format provides information on handling big data in a systematic way providing the learner a perfect platform to understand big data and its applications in real world.

Contents

Task 1

Introduction

Task 1.1

Exploratory Data Analysis of Weather Data

Relationship between the variables using Scatter Plot diagram

Chi Squared Test Analysis and Output

Top five significant variables

Task 1.2

Introduction

Decision Tree Rules

Conclusion

Task 1.3

Weka Logistic Regression

The Logistic Regression processes

Output of Regression

Odd Ratios

Task 1.4

Model Validation

Decision Tree Validation

Decision Tree Accuracy

ROC Curve for Decision Tree

Lift Chart for Decision Tree model

Logistic Regression Validation

Logistic Regression Accuracy

Confusion Matrix

ROC (Receiver Operating Characteristic) analysis for Logistic Regression

Lift Chart for Logistic Regression Model

Conclusion

Task 2

Introduction

Task 2.1

Enterprise Data Warehouse (EDW) Architecture of a water utility company

Task 2.2

Component description of the proposed architecture

Conclusion

Task 2.3

Security Privacy and Ethical Concerns when using Big Data platform

Task 3

Tableau Desktop

Introduction

Task 3.1

Impact of wildlife strikes with aircraft over time for a specific origin state

Task 3.2

Wildlife Strikes by flight of phase and time of the day

Task 3.3

Damage Caused by Wildlife Species to Aircrafts

Task 3.4

Damage caused to Aircraft in different state

Task 3.5

Interactive Dashboard on Wildlife Strike

List of References

Task 1

Introduction

RapidMiner is a software platform for data mining. This software is used for data preparation, machine learning and predictive model deployment. Predictive analysis is done to know if there will be rain tomorrow using the data provided in the dataset.

Task 1.1

Exploratory Data Analysis of Weather Data

Process: To start with the EDA process, the first step is to upload the data set to the rapid miner and then connect its output node to the resulting node of result.

Fig 1 EDA process diagram

Findings of EDA Analysis

Attributes	Min/ Least	Max	Average	Standard Deviation	Most Frequent	Missing Values	Inconsistency
MinTemp	-8.5	33.9	12.16	6.394		1665
MaxTemp	-4.8	48.1	23.19	7.135		1465
Rainfall	99.2				0 (86311)		NA (3530)
Evaporation	86.2				NA (58198)
Sunshine	14.5				NA (64534)		NA (64534)
WindGustDir	NNE (6260)				NA (9871)		NA (9871)
WindGustSpeed	6				NA (9980)		NA (9980)
WindDir9am	WSW (6689)				N (11339)		NA (9980)
WindDir3pm	NA (3884)				SE (10227)		NA (3884)
WindSpeed9am	83				9(12844)		NA (1728)
WindSpeed3pm	87				17 (11917)		NA (2762)
Humidity9am	0	100	68.676	19.038		2821
Humidity3pm	0	100	51.463	20.808		4891
Pressure9am	980.5	1041	1017.56	7.10		14603
Pressure3pm	977.1	1039.6	1015.16	7.03		14567
Cloud9am	9				NA (52804)
Cloud3pm	9				NA (55717)
Temp9am	-7.2	40.2	16.983	6.498		1997
Temp3pm	-5.4	46.1	21.657	6.950		3441
Rain Today	NA (3530)				NO (104498)		NA (3530)

Relationship between the variables using Scatter Plot diagram

Fig 1.1 Relation between Wind Gust Speed and Wind Direction

Fig 1.2 Relation between WindSpeed at two different time of a day

Fig 1.3 Relationship between Evaporation and WindDir3pm

Chi Squared Test Analysis and Output

Fig 1.4 Variables for Analysis

Top five significant variables

The top five variables chosen for this analysis are Evaporation, WindDir3pm, WindDir9m, Windspeed3pm, WindGustDir

These five variables are selected based on weight by Chi Squared Statistical tool.

Evaporation is the process of converting water in vapor. When evaporation rate is high, there is higher chance of rain as the atmosphere is loaded with humidity.

Wind Direction at different times of the day is also important to determine the chance of rain. As the weather data set has cities from Australia which is a largest island continent, the wind direction is important parameter to determine the possibility of rain.  If the wind has passed over sea or any large river, it will pick up more moisture, which could bring rain. If the wind has passed over dry, hot land terrains, the air will be hot and dry and the chance of rain is nil. Wind direction changes caused weather changes.

Windspeed is the result of the difference in air pressure. The lesser the air pressure, the greater the windspeed.  Windspeed is measured in kilometres per hour as per Australian Government Meteorology Department and is expressed in knots for aviation use.

Wind gust can be described as the sudden increase in the windspeed followed by a lull. Information on the wind speed and direction provides clues to predict weather.

Task 1.2

Introduction

Decision Tree Model machine learning technique is carried out on the clean weather dataset. The five variables discussed in task 1.1 is selected and the Decision Tree process is run.

Fig1.5 Decision Tree Process

Fig 1.6 Decision Tree Diagram

Decision Tree Rules

Fig 1.7 Decision Tree Rules

Conclusion

The decision tree model predicts there will be no rain tomorrow based on 138307 examples of the selected attributes, the least number of examples predicted Yes for rain tomorrow is 30283 and the 108024 examples were predicted NO for rain tomorrow.

Task 1.3

Weka Logistic Regression

Logistics Regression is regression analysis tool used to conduct the analysis on the given weather data set, the numerical attributes are converted to polynomial and then to binominal.

The Logistic Regression processes

The clean data set is used.  Select attribute operator is used to select the top five attributes (on basis of chi square). Then Nominal to Binomial operator is used to convert the nominal values of attribute selected to the binomial as logistic regression supports the binomial attributes. Set role operator is used with Rain Tomorrow as label attribute. Then the logistic regression model is used and its mode, example set, weight and threshold node are being connected with the resulting nodes of process to view the predictions of logistic regression model.

Fig 1.8 Logistic Regression Process

Output of Regression

Fig 1.9 Result of logistic regression model

Odd Ratios

Fig 2 Logistic Regression Ratios

Task 1.4

Model Validation

Decision Tree Validation

Fig 2.1 Cross Validation Process

Decision Tree Accuracy

Fig 2.2 Decision Tree Accuracy

ROC Curve for Decision Tree

Fig 2.3 ROC curve

Lift Chart for Decision Tree model 

Fig 2.4 Lift Chart Result

Logistic Regression Validation

Fig 2.5 Cross Validation Process

Logistic Regression Accuracy

Fig 2.6 Model Accuracy Result

Confusion Matrix

Fig 2.7 Confusion Matrix

ROC (Receiver Operating Characteristic) analysis for Logistic Regression

Fig 2.8 ROC Analysis Output

Lift Chart for Logistic Regression Model

Fig 2.9 Lift Chart

Conclusion

Based on the above analysis, the logistic regression model is more appropriate in the prediction of rain tomorrow.  The model gives a large number of sample set 108024 as true negative which indicates it will definitely not rain tomorrow.

Task 2

Introduction

The state-owned water company manages and provides water services to 1.3 million consumers in the industrial and domestic area.  As it is a state-owned company it is the largest distributor of water delivering 0.4 million cubic meters of water per day through a water network length measuring 5000 KM and 2000 km of connections having more than 100,000 fittings, valves and controls.  There are 34 pumping stations, 180 medium and large reservoirs and tanks. The company process 0.10 million cubic meters of waste water per day, Waste water network length is 2000 KM.

Task 2.1

Enterprise Data Warehouse (EDW) Architecture of a water utility company

The role of the EDW is to integrate the data across the operational system which may be in the operational silos as well as distributed geographically. The EDW of this water utility company includes the large volumes of data sources, data process systems, data analysis and reporting, data ware house, data marts, big data analytical applications and tools.

   Data Flow

   Process Flow

Task 2.2

Component description of the proposed architecture

Data Sources: Huge amount of data is generated real time in water utility company from sensors and meter readings, telemeters which is a part of data from IoT.  This data is part of Online Transactional Processing System. An example of Real-time data sources in water utility industry is water pressures data that is collected every 10 seconds and which is refreshed every 30 seconds.  Data from sensors and actuators also constitute real time data.  Customer information, their interactions, water usage, new connection requests, billing, customer complaint data from consumer information system (CIS) and the data collected throughout the life cycle of the customer is part of the Customer relationship management CRM data.

ERP data of the state utility company includes data from the Human Resources department, Supply chain management data like water quality, work orders, raw water in reservoirs and canals, water distribution, rainfall data, sales and marketing data, supervisory control and data acquisition (SCADA) system, regulatory data in a laboratory information management system (LIMS), data from surveys, social media data, data about losses due to leakage, grid data, data on Incorrect billing (inaccurate metering, data handling errors), Frauds, evaporative losses, waste water generation is also a part of ERP systems. There is also the Historical Data about the past events and Asset management data of the water utility company which is data about the pipes, pumps, their maintenance information, inventory data, manuals, diagrams, standards, references, Operating and maintenance costs, Revaluations, Warranty and Insurance, Budgets. Data from Remote Sensing, Geographic Information Systems, Global positioning system, sensor networks, scientific research projects are some of the data sources that are very unique to the water utility industry. Unstructured data such as emails, instant messages, images, videos and social media post is also a source of data for the water company.

Data Storage: Data is stored on-premises in a Hadoop Distributed File System (HDFS) that can hold big volumes of large files in different formats. In addition to this, the company also uses Cloud Services to store its data. Cloud storage vendor Windows Azure Storage Blob (WASB) is providing support to company by offering endless storage space for the huge data in its native format and this can be scalable at any time. Some other examples of cloud service providers are Microsoft ADLS, AWS S3 and Google cloud storage.

Extraction Transformation Load Process: Large volume of data generated from connected devices and sensors is captured, stored and processed using Hadoop. Hadoop combines data from various data sources as discussed above, which is in different formats. Apache NiFi recognizes all of those sources and moves the data to a central repository for storage and analysis both in real-time and batch processing.  XML allows integration of data from data source to relational databases by tagging the data right when the data is created or at a later stage.  This integration allows end users to access to information from the data. For example, data from GIS-based applications is integrated in XML format to provide a unified view which can accessed on web services.  In the staging process the data is stored in NoSQL database.

The data from sensors and other telemeter devices is streaming at a high velocity and it is data in-motion. This real-time data ingestion in the architecture is performed by Microsoft Azure Event Hubs, Kafka. These softwarecaptures and store the data in a folder for stream processing by acting as a buffer for message ingestion, message queuing.

One example of Real-time data processing use in the water utility company is to monitor Water Quality parameter profiles like the concentration of Nitrate over time to determine the travel time between sites. This will show the water age in real-time. The information from this can be also be used to verify hydraulic model.

Batch Processing of data is performed to generate daily, weekly, monthly reports in the utility company.  Big data files are processed to prepare these reports. The solutions offered by Azure Data Lake Analytics use Hive, Pig, MapReduce in the Hadoop cluster to read the source data, filter, aggregate and prepare the data for analysis.

Stream Processing: After ingestion of real-time data Apache streaming technologies like Storm and Spark Streaming in used in the Hadoop cluster for the stream analysis of data. SQL queries are run on the large set of data to prepare data for analysis. The data can be used to generate ad hoc reports, helps to determine the unseen relationships and causes like fraud detection, help to establish KPIs and benchmarking process to run the water plant.

Machine Learning: The information hidden in the data is discovered by statistical analysis. Some example of machine learning is Classification, Validation, Supervised and unsupervised methods of machine learning. Neural Network model is a machine learning technique used by water utility industry to assess water quality in water distribution network, or the river flow forecasting, water demand prediction.

Data ware house: Data ware house is the federated repository for the data collected by the number of operations in the ELT stage. It is a single source of truth. The technology used in Big Data warehouse is Hadoop MapReduce and Hadoop YARN as execution engines, the Hadoop Distributed File System (HDFS), and HBase as replacement for BigTable. Cassandara is a distributed data management system, Zookeeper is a high-performance coordination service for distributed applications, Pig and Hive for data warehousing, and Mahout for scalable machine. Pig is used as a scripting platform and Hive adds structure to the data as it is a simple SQL language used to query data.

Data marts:This component of the architecture is the subset of the data warehouse and the structure and pattern of these data marts is specific to the data ware house environment and these data marts are primary used by regional offices of the utility company. The data marts retrieve the data which maybe a requirement of a single functional area of a specific department of the utility company.

Data Orchestration: Orchestration technology such Apache Oozie and Sqoop is used in big data architecture to automate processes like repeated data processing operations, to encapsulate workflows, data transformation, data movement between multiple sources, sinks and loads the processed data into an analytical data store, or push the results straight to a report or dashboard.

Analysis and reporting: The vast amount of data which runs is petabyte is analysed using data modelling layer, such as a multidimensional OLAP cube or tabular data model in Azure Analysis Services. It also supports self-service BI, using the modelling and visualization technologies in Microsoft Power BI or Microsoft Excel. Analysis and reporting is also in the form of interactive data exploration by data scientists or data analysts. Visualization tools like Tableau, Jaspsersoft, Qlik, Power BI, open source like GIS/ spatial, plot.ly help to generate interactive dashboards to analyse the hidden data, providing meaningful reports across the organization. These tools help to visualize near-real time data to check billing irregularities, service areas, ad hoc reports for middle management team.

Data Orchestration: Orchestration technology such Apache Oozie and Sqoop is used in big data architecture to automate processes like repeated data processing operations, to encapsulate workflows, data transformation, data movement between multiple sources, sinks and loads the processed data into an analytical data store, or push the results straight to a report or dashboard.

Conclusion

The proposed big data architecture will boost the company performance as it will help the company to manage its big data in an efficient way. Batch processing can be performed on data sources at rest, for data in motion real-time processing will give the needful insights. The available data can be explored in many ways for predictive analysis and machine learning techniques can be applied to oversee the operations of the company.

Task 2.3

Security Privacy and Ethical Concerns when using Big Data platform

The key security, privacy and ethical concerns associated in water utility company called Watercare is discussed below. This company collects, store, and analyse large quantities of data about customers’ locations, billing, online transactions, usage patterns, interests, demographics, and more.

Security Recognizing and specifying the exact location of the processing data is a security problem and leads to regulation breaches.  This is primarily because the big data concept is based on parallelism and large amount of data is stored in random distribution in different clusters. Cyber security breaches have known to occur where hackers breached the security of the Watercare company and tampered with critical systems to control water flow. The Watercare company has data worth more than $40 million which the cyber-criminal is trying to exploit. The access to this information can lead to the theft or loss of data and damage to internal business systems and customer-facing platforms. The cyber-attack inhibits the company from conducting business as its communication is shut down. Another threat is the business email scam, whereby a criminal pretends to be a from the company directing customers to transfer funds electronically to a seemingly legitimate account which is actually not. Hackers can also remotely gain access to water pumps and manipulate the movement and treatment of the water to cause disruption to the service. The security breaches can be financially motivated, also, there is greater possibility of an angry or dissatisfied employee, customer or even competitor, having the capability to disrupt operations through a cyber-attack.  Watercare company has made the cyber security awareness programme compulsory in its company in order to equip employees with better understanding about cyber security.  The call centre staff, field engineers, operations teams engage with external networks on a daily basis. These people are trained to identify cyber security threats and take it as their responsibility to flag it and to know to how manage that risk. Watercare company keeps access to their SCADA systems very secure to avoid security breaches. Technicians, engineers, and operational decision-makers are given access to SCADA network based on their work role profile. No access is given to this system just for the convenience sake for convenience. This reduces vulnerabilities to cyber-attacks. Watercare company is very protective of the information that goes of its company. Steps are taken to reduces vulnerability to cyber-attacks by not positing any detailed information of its utility consultants and contractors, employee names and email addresses, treatment facilities or any other information related to its assets, description of any of its projects involving large capitals.

Privacy is a major issueassociated with the big data analytics. Big data technologies are to assign the sensitive data and the current big data analytics is responsible for providing the data storage and processing facilities with the same priority and do not associate with special actions such as bling processing and encryptions of data. Therefore, if the hacker or the malicious node gains the access to the clusters it would be easy for the hacker to steal, exploit or alter the contained records for malicious purpose. Watercare company also collects customer data for its research purpose in form of questionnaires, behaviours, experiments, to do microtargeting, and service customization. This data of the customers poses a greater risk as the personal information may be disclosed, misused, or used in ways that will adversely affect them in the future. For example, risks of financial loss, identity theft.

Watercare company has adopted notice and consent mechanism to address the privacy issues. The de-identification technique is also applied according to terms of service agreement. For research purposes, the company implemented an extensive review process utilizing privacy controls and explicit and informed consent forms, controlled statistical disclosure, and data use agreement.  There is procedure control to limit access to data and its use.  The company follows state laws to protect data privacy which is Privacy Act 1993 (New Zealand).

Ethical Concerns: One example of ethical concern that can discussed in water utility company is huge amount of customers GPS data. The issues that can raise is location-based stalking.  There are remedies taken by the company for data protection and to safeguard the interest of customers and the society in whole.  The Watercare company collects data and stores them on internal servers protected by firewalls and this information is accessible only to authorised staff. The company does not involve trade customer information with third parties to gain monetary benefit.  There are many third parties’ websites listed on the company website and if the customer is accessing those website, Watercare has notified its customers to understand the privacy policies of each linked website.

Task 3

Tableau Desktop

Introduction

Tableau is an interactive visualization tool helpful in data analysis, thereby giving meaningful insights about the data which is used by the decision makers to run their businesses. Aviation Wildlife Strike data set will be analysed to find the hidden information in the data and an interactive dashboard will be created.

Task 3.1

Impact of wildlife strikes with aircraft over time for a specific origin state

Fig 3.1 Wildlife Strike

To perform this analysis bar chart is used and colour legend is assigned to the Airline operator. The other attributes used in this analysis are Origin state, Impact to flight, Time of day, Aircraft Type and number of wildlife strike, number of records found. For analysis of the impact of the wildlife strikes the two states chosen is Texas and California.

Origin State Texas: In the analysis the maximum number of wildlife strike was one for Southwest Airlines where 920 records. This strike occurred to Airplane during day time, although there was no damage caused.  Another observation is on the Military aircraft where the wildlife strike was in the night, and the aircraft had to do precautionary landing.

California: In the analysis the maximum number of wildlife strike was one for Southwest Airlines where 359 records are during night time. This strike occurred to Airplane during night time, although there was no damage caused.  Another observation is on the Business Airlines where the wildlife strike was in the night where there were 36 records and one life strike happened and the impact to the aircraft was none.

Task 3.2

Wildlife Strikes by flight of phase and time of the day

On analysis of the flight phase, the maximum number of strikes to the aircraft is during the Approach phase. During this phase the flight is within specified airspace and is waiting for further instructions. This increases the chance of bird strike with the aircraft as birds are common on this altitude. One observation on Southwest Airlines during day time, there was one bird strike have 390 records.

Fig 3.2 Strike during phase of time

Task 3.3

Damage Caused by Wildlife Species to Aircrafts

There a number of bird species moving in the airspace that has caused significant damage to the aircraft monetarily. The most significant damage was caused by unknown bird species of medium size that caused damage totalling $45,329,176.

Fig 3.3 Damage to Aircraft

Task 3.4

Damage caused to Aircraft in different state

The dataset has large number of examples that does not belong to the USA country. Filter examples is used to remove these examples. The damage caused by wildlife strike is mentioned in the analysis. The maximum damage is observed in New York for US Airways, where the total cost of damage was $37,948,803 in year 2009.

Fig 3.4 Damage caused to A(n.d.)ircraft in different origin state

Task 3.5

Interactive Dashboard on Wildlife Strike

This interactive dashboard gives information about the wildlife strike occurring in different state, strikes that happened during the which phase of flight, the specific wildlife species involved in this damage and the cost to the airlines in different origin state.  All of this information is contained in a single screen giving the user the freedom to move the cursor around the graphs to find more details.

Fig 3.5 Dashboard

Conclusion: This information is helpful to aviation industry can reduce the damage to their aircraft by altering its flight operations and develop more effective prevention strategies.

List of References

Altman, M, Wood, A, O’Brien, DR & Gasser, U 2018, ‘Practical approaches to big data privacy over time’, International Data Privacy Law, vol. 8, no. 1, pp. 29–51.

‘Bangalore Water Taps IBM for Big Data Analytics’ 2014, accessed October 8, 2018, from <https://www-03.ibm.com/press/in/en/pressrelease/43242.wss#resource>.

‘Bird Strike Reporting – SKYbrary Aviation Safety’, accessed October 17, 2018, from <https://www.skybrary.aero/index.php/Bird_Strike_Reporting>.

‘Cyber Security: How Water Utilities Can Protect Against Threats’, accessed October 15, 2018, from <https://www.waterworld.com/articles/wwi/print/volume-32/issue-1/technology-case-studies/cyber-security-how-water-utilities-can-protect-against-threats.html>.

‘Druid |’, accessed September 26, 2018, from <http://druid.io/docs/latest/design/>.

‘Druid | Introducing Druid: Real-Time Analytics at a Billion Rows Per Second’, accessed September 26, 2018, from <http://druid.io/blog/2011/04/30/introducing-druid.html>.

‘Druid | Use Cases’, accessed September 26, 2018, from <http://druid.io/use-cases>.

‘EB7306_IoT_Analytics_Architecture_Time_Series_White_Paper.pdf’, accessed October 8, 2018, from <http://assets.teradata.com/resourceCenter/downloads/AnalystReports/EB7306_IoT_Analytics_Architecture_Time_Series_White_Paper.pdf>.

‘How Dashboards Helps Decision-Makers at Water Utilities’ Valor Water Analytics, accessed October 7, 2018, from <http://www.valorwater.com/blog/2017/7/25/how-dashboards-helps-decision-makers-at-water-utilities>.

‘Manage Data-at-Rest and Deliver Big Data Analytics with Hortonworks Data Platform (HDP)’ Hortonworks, accessed September 25, 2018, from <https://hortonworks.com/products/data-platforms/hdp/>.

Martin, KE 2015, ‘technology, and identifies the ethical issues it faces. These issues arise from reselling’, , p. 19.

‘Meeting the Demand for Data-driven Change in the Utilities Sector’ , p. 12.

‘New Zealand’s official weather forecasts & warnings – Meteorological Service of New Zealand (MetService)’, accessed October 10, 2018, from <https://www.metservice.com/national/home>.

Nikolajeff, J-P ‘Analysis of the Bird Strike Reports Received by the Finnish Transport Safety Agency between the Years 2000 and 201’, , p. 60.

Nowalk, M ‘Precipitation and evaporation’, , p. 7.

Research Ethics and New Forms of Data for Social and Economic Research 2016, accessed October 16, 2018, from <https://www.oecd-ilibrary.org/science-and-technology/research-ethics-and-new-forms-of-data-for-social-and-economic-research_5jln7vnpxs32-en>.

Shahrivari, S 2014, ‘Beyond Batch Processing: Towards Real-Time and Streaming Big Data’, Computers, vol. 3, no. 4, pp. 117–129.

‘The wind across Australia’, accessed October 2, 2018, from <http://www.bom.gov.au/climate/averages/wind/>.

‘Veolia_VIC_Waternamics_Brochure_V5.pdf’, accessed October 4, 2018, from <https://www.veolia.com/anz/sites/g/files/dvc1131/f/assets/documents/2017/07/Veolia_VIC_Waternamics_Brochure_V5.pdf>.

‘Watercare – Privacy policy’, accessed October 7, 2018, from <https://www.watercare.co.nz/Privacy-policy>.

‘WIND DIRECTION AND WEATHER FORECASTING’, accessed October 2, 2018, from <http://www.theweatherprediction.com/habyhints2/432/>.

ZoinerTejada ‘Big data architectures’, accessed September 29, 2018, from <https://docs.microsoft.com/en-us/azure/architecture/data-guide/big-data/>.