ML vs Real-World Problems

In the last few years, the explosion in the field of ML and AI have been significant for many startups to disrupt the traditional markets.

Machine Learning and basic programming can never substitute each other but machine learning techniques can be used when traditional programming can’t deal efficiently with the problem. For example, a simple weather forecasting system, the system interface, data preprocessing, data visualization and so on will be implemented in a usual programing language(Java, Ruby, Python..) but the forecasting part will require ML models.

Machine Learning needs data and with an abundance of data present on the internet, one can process it, manipulate it, to create a model.

Machine Learning usages are abounding. They make up the core or difficult parts of the software you use on the web or on your desktop every day. Think of the “do you want to follow” suggestions on twitter or the speech understanding in virtual assistants.

I would like to share some impressive real-world examples of applications of machine learning(ML) which I recently came across.

Medical diagnostics for detecting diseases: Doctors and hospitals are now increasingly getting assisted in detecting diseases like skin cancer faster and more accurately. A system designed by IBM correctly picked the cancerous lesions(damage) in the images with 95% accuracy where a doctor’s accuracy is usually between 75% — 84% using manual methods.

I and my friend also published a paper regarding this issue our main purpose was-. About 1.7M new breast cancer cases were diagnosed in 2012. As of 2018, nearly 12.4% of women in the US are expected to develop invasive breast cancer over their lifetime. Mammography has always been the most effective technique for the screening of breast cancer. But, the low positive predictive value of breast biopsy which results from the interpretation of mammogram leads to nearly 70% unnecessary biopsies with benign outcomes. To solve this problem, supervised machine learning classification algorithms can be applied to develop a machine learning models which can predict the rigorousness of a mammographic mass with the help of BI-RADS attributes and the patient’s age. The result was as follows -

Content (image, video, text) categorization: ML-based computational approaches largely solves the tedious task of classification of huge amounts of documents based on their types. For example, in a library, the books can be classified as fiction, literature, scientific, etc. without requiring a person to go through all the book titles. The link given below takes an image/video, reads it, predicts the related tags and gives similar images/video. Thus classifying the image/video based on its content. Try this once to check if it works correctly or not. Visual recognition API and services

Use of NLP(Natural Language Processing): Some other examples of the application of ML are in natural language processing for sentiment analysis, Email spam detection, targeted advertisements(Google Ad-sense), recommendation engines used by E-commerce sites, pattern mining for market basket analysis.

Suspicious activity detection from CCTVs: The basic idea is to train a model that will learn and analyze CCTV video and normal activities all the time and if anything suspicious happens it would alert authorities. http://www.wisdom.weizmann.ac.il/~vision/Irregularities.html

Originally published at https://www.linkedin.com.