Decision Trees, Regression Trees, Classification Trees, Boosted Trees, Rotation Forest, Random Forest.. To see the machine learning woods for the trees we are giving you an overview over the most important terms:
Data can be divided into structured and unstructured data. Structured data for instance can be sensor data saved as text files or sales measures from a datawarehouse. On the other hand, images and speech are unstructured data.Both data types can be analyzed with supervised or unsupervised machine learning techniques.
Machine Learning (ML)uses statistics, computer science and artificial intelligence to extract knowledge from data, with disruptive results in the last couple of years. Using ML techniques to find patterns in big data is also called data mining. Classical coding usually tries to translate a conceived model into code, which puts input into output. ML switches this approach: the machine "learns" the model on it`s own, just via comparing input data with output data. "Machine learning is the science of getting computers to act without being explicitly programmed" (Andrew Ng). The benefit of this approach is that even new inputs can be correctly labeled by the system without any explicit rule coding beforehand.
The idea behind supervised learning is, that there are already correct answers to these data from the past. For instance, you might have already generated sales in the past and now you want to extrapolate your sales data into the future using simple linear regression. Instead, if you want to categorize images of your production process into scrap and yield you can use logistic regression. Categorizing a risk into more than two classes you enlarge logistic regression multinomial. Product recommendations for your customers are also quite frequently used, in this case using matrix factorization.
Use Python and R Frameworks to find patterns via clustering your multidimensional data or isolating extraordinary datapoints. Contrary to supervised learning your data must not be labeled when trying to gain insights out of it via unsupervised learning.
Transfer Learning means to use pretrained models which can seriously lower effort, time and costs while training. This can be a fruitful approach for computer vision projects, because instead of starting from zero you can rely on successfully pre-trained models for similar image objects. This can also make a lot of sense if you only should have a limited amount of training pictures.
The possibilities on the field of machine learning are diverse and development is face-paced. No matter if your company is already staffed with Analysts, Data Engineers, Data Scientists, Chief Data Officers and Statisticians or if you`re right at the start: use the chance offered by the "sexiest job of the 21st century" (Harvard Business Review: Data Scientist) because:
“By 2020, some 50 billion smart devices will be connected, along with additional billions of smart sensors, ensuring that the global supply of data will continue to more than double every two years” (McKinsey Quarterly: Straight Talk About Big Data). But the real surprise about this McKinsey survey is that nowadays only about 1% of all this data is estimated to be analyzed at all. To change that fact is our mission.
If you are looking for support for the sake of discovery, interpretation and communication of meaningful patterns in your data we look forward to hearing from you. We are here to help you gathering new insights from your data via spotting unknown trends, seasonality or patterns. We offer both analysis as a service as well as trainings.On the right medium you will successfully culture seed for Data Analysis. Please take a look with us on the Data Science "petri dish":
Benefit from our expertise! We are looking forward to getting to know you!