Deep Learning in Binary Classification Examples
A fascinating look into a few important applications of binary classification.
Binary classification for deep learning can come in many shapes and forms, such as having a model distinguish between cats and dogs, mean and nice Twitter tweets, positive and bad reviews, and even more. Though it may seem simple and straightforward, binary classification can be used in numerous types of fields and can even serve as critical solutions.
Spam, scam, and dangerous mail/message filtering
As the digital world continues to grow, so can the prevalence of spam, scam, and dangerous messages sent. With enough data containing harmless and harmful messages, a deep learning model can be trained to distinguish between good and bad text and notify users before they click on/reply to a person. This can be especially helpful for children, the elderly, and others who are new to technological communication and can help reduce the number of people falling prey to scam or malicious messages.
Financial and account fraud detection
As the digital world continues to grow, so is the prevalence of payments and transactions using technology such as PayPal and communications such as using email. For example, a model can be trained to find unusual patterns in financial transactions that may indicate fraud, such as irregular or strange credit card transactions, transferring large sums of money to another account, etc. or find unusual patterns in account login behavior, such as logging into an email account in multiple different locations. For example, Gmail already has this feature where they will detect unusual logins if we log into our Gmail account in a different location, such as on vacation. This can be used to secure our accounts for platforms such as PayPal, email, and more and prevent malicious users from using our accounts for harm.
Distinguishing between cancerous and noncancerous skin conditions
Most moles are harmless, but certain types can be dangerous and cancerous. For example, following the ABCDE rule, moles that are asymmetric, have irregular borders, multiple colors, larger than the diameter of a pencil, eraser, and evolves/changes over time can be signs of melanomas, a common type of skin cancer. In this case, if we train a model on images of normal moles and skin conditions and cancerous ones like melanoma and basal cell carcinoma, it can possibly be used as an alternative solution for people to conveniently check at home if any current skin conditions they have are dangerous or not. This not only saves time from going to frequent doctor visits to check but also saves money since doctor visits can be expensive in many places.
Using miRNA biomarkers to detect cancer
It has been recently discovered (around 10–20 years ago) that miRNA can serve as important biomarkers of cancers. Feeding miRNA data into models can help them train to identify the specific sequences that indicate cancer and ones that don’t, hence becoming a task of binary classification. For example, the GDC Cancer Portal has data for cancers such as lung, breast, kidney, and more that all contain unique miRNA sequences for a variety of patients that can be used to train models. This can serve as an secondary solution to check if a person has cancer or not to provide an early diagnosis before it’s too late.
Caveats
Of course, these ideas have not been widely integrated as general solutions yet, but they can serve as alternative solutions if they are tested and developed more. As deep learning continues to grow, who knows what other inventions could be created from it. :)
Sources Used
https://vitalflux.com/classification-problems-real-world-examples/
