Client Background
Client: A Leading Tech Firm in the USA
Industry Type: IT Consulting
Services: Software, Consulting
Organization Size: 100+
Project Description
We need to create a notebook with solutions to binary classification-related anomaly detection problems. We need to use machine learning and deep learning models which have greater than 90% accuracy.
Our Solution
We created a notebook for anomaly detection. We used 10 to 15 machine learning and deep learning models but only 3 different types of auto encoder models that were giving greater than 90% accuracy. We trained all 3 models on one classification data which have anomalies and evaluated trained models on test data.
Project Deliverables
A notebook that has solutions for anomaly detection related classification problems and accuracy should be above 90%.
Tools used
Google colab notebooks, Tensorflow, Google drive
Language/techniques used
Python programming language, Machine learning, Deep learning, Data analysis and Data visualization.
Models used
Auto Encoder and Variational Auto Encoder
Skills used
Python, Data Analysis, Data visualization, Machine learning, Deep learning.
Databases used
MS Excel
What are the technical Challenges Faced during Project Execution
- Most of the anomaly detection models work with regression type data and this problem was classification problem so we need to deal with classification data.
- Getting high accuracy is also a tough challenge for us because there are only a few models which work well on anomaly detection related classification problems.
How the Technical Challenges were Solved
- So we have limited models for this problem so we used only classification models like Autoencoders, Isolation forest and one class svm.
- Only Autoencoder was giving high accuracy so we worked with different types of autoencoders like variational autoencoder and normal autoencoder.