Specialisation Foundation → Specialist
Performance Data Science
The statistics under every performance number: describe a ship's signals honestly, clean them without inventing anything, find what actually moves together, and build a model you are willing to defend.
- Courses
- 4
- Lessons
- 32
- Time
- About 6 hours
- Price
- €559+ VAT
- Assumed knowledge
- No statistics beyond school level, and no programming. The methods are taught in prose and most are worked in numbers on the page, so the sequence is followable without running a line of code, and most of what it teaches can be reproduced with standard spreadsheet functions; the heavier machinery, the Kalman filter, principal component analysis and the neural network, is explained for what it buys and costs rather than computed by hand. Our course Fleet Data & Performance Monitoring is the ideal companion: it owns the data estate this sequence analyses.
€559 + VAT for the sequence
The 4 paid courses cost €736 bought one at a time; as a specialisation they are €559, a saving of €177 (24%).
Buying online needs JavaScript switched on in your browser. Otherwise write to info@blueautonomy.gr and we arrange access.
You own this specialisation. Every course in it is open to you.
Every performance claim in this Academy ends in a number, and every one of those numbers came out of a pipeline that somebody chose: a filter threshold, a smoothing window, a correlation, a fitted curve. This sequence teaches those choices as a discipline rather than as software defaults. It works throughout on shipboard signals (speed through water, shaft power, propeller revolutions, fuel consumption, generator load), and it is honest about the thing that makes marine data hard: the ship never repeats a condition, so the sample you have is never the sample you wanted.
What you'll learn
- Describe a shipboard signal in the measures that suit it, and recognise when the mean, the standard deviation and the correlation coefficient are all quietly lying
- Choose an outlier test on purpose: z-score, IQR, MAD, Chauvenet, regression residual or Mahalanobis distance, and state what each one assumes
- Smooth a noisy signal without erasing the event you were looking for, and defend the window you picked with variance reduction and RMSE rather than taste
- Separate real dependence from coincidence and confounding, using rank, distance and partial correlation where Pearson misleads
- Compress a many-channel telemetry stream to the handful of modes that carry its energy, and read the loadings back as physics
- Judge a fuel-consumption model the way its consequences deserve: physics where physics is known, data where it is not, and whether a specific decision should be taken on its output, given the payoffs and the model's real assumptions
The sequence
Each course stands alone; this order is the order the knowledge builds.
How the courses build on each other
The order is the order of the pipeline itself, and each course earns the next. Describing Vessel Data teaches what a signal is before anyone touches it, because every later decision (which outlier test, which correlation coefficient, which model) is decided by the shape of the distribution you are holding. Cleaning the Signal then does the touching: outlier detection and smoothing, taught as a pair because they are the two ways to be wrong about the same point. Finding Relationships in Performance Data asks what moves with what, and spends most of its time on the ways that question is answered falsely. From Physics to Models closes the sequence by building something that predicts, and by drawing the line between a model that is accurate and a model that is trustworthy. Two boundaries are drawn on purpose. Fleet Data & Performance Monitoring owns the data estate (where the signals come from, what governs them, what architecture holds them), and this sequence owns the methods applied to it. The two are companions, and the ground they share (flagging a suspect record rather than deleting it, keeping the impossible apart from the merely improbable, and reporting what survives cleaning beside what was logged) is stated in both because each depends on it. Our ISO 19030 course owns the standard's filter chain as a written procedure, thresholds and reference periods as the standard frames them; here you learn the statistics underneath that chain, which is what you need on the day your data does not qualify and the standard stops answering.
-
Paid Foundation
Describing Vessel Data
How to look at a shipboard signal before you do anything to it: the measures that describe it honestly, the plots that expose what the measures hide, and the normality question that quietly decides which methods you are allowed to use next.
-
Paid Practitioner
Cleaning the Signal
Outlier detection and smoothing taught as one subject, because they are the two ways to be wrong about the same record: how to find the values that should not be trusted, how to quieten a noisy channel without erasing the event you were looking for, and how to defend both choices to someone who disagrees.
-
Paid Practitioner
Finding Relationships in Performance Data
What moves with what, and how often that question is answered falsely: correlation and its rank, distance and information-theoretic alternatives, the confounders and mediators that make the same arithmetic mean opposite things, and the decomposition that compresses a many-channel telemetry stream to the few modes carrying its energy.
-
Paid Specialist
From Physics to Models
Building a predictive model of a ship that someone should be willing to act on: what physics contributes that data cannot, which variables are inputs and which merely emerge, why a time-series split is the honest test, what a neural network buys and costs, and how to attach an uncertainty that survives contact with a decision.
The specialisation certificate
Every course above ends in a graded final exam and a certificate. Hold the certificate for every course, and a capstone project unlocks: one scenario that draws on the whole sequence, marked on our server. Pass it and you get a second, separate certificate for the specialisation itself, naming every course it comprises: its own number, verifiable the same way.
Who teaches this
Common questions
Do I need to know Python?
No. Every method is taught in prose, most are worked through on numbers you can follow with a calculator or a spreadsheet, and each is named precisely enough to implement in whatever tool you already use. Nothing in the lessons, the exams or the capstone depends on writing code.
Is this a general data-science course with ships bolted on?
The opposite. Almost every example is a shipboard signal, and the few that are not are textbook cases kept because they are hard to argue with. The recurring lesson is that marine data breaks the assumptions general courses take for granted: the conditions never repeat, the interesting values live in the tails, and the variable you most want is the one nobody measures. The methods are standard; the judgement is not.
How does this relate to ISO 19030?
ISO 19030 hands you a filter chain and a set of thresholds, and our ISO 19030 course teaches them as the standard frames them. This sequence teaches the statistics a chain like that rests on, the outlier tests and the steady-state test above all, and what to do when a ship's data does not satisfy it. The two are meant to be read side by side for that reason.