Prediction and Its Impact on Its Attributes While Biasing MachineLearning Training Data
Jaggannagari Kavitha, Jangala. Sasi Kiran, Srisailapu D Vara Prasad, Krushima Soma, G Charles Babu, Sreenidhi Sivakumar · 2022
Machine learning models are built utilizing biased training data that comes from human experience. The data gathered reflects the cognitive bias that human's display in their actions and thought processes. The most effective machine learning models are considered to resemble human cognition; as a result, these models are biased. For improved explainable models, bias detection and evaluation are crucial. This study identifies bias in learning models in relation to cognitive bias in humans and suggests a cutting-edge method for identifying and evaluating machine learning bias. It also tries to identify bias in the dataset. The potentially skewed qualities are seen in the deployed dataset. Prior to employing the notion of alternation function to swap the values of PBAs and analyze the influence on prediction using KL Divergence, we first choose a few common biased characteristics. Compare the KL divergence values from the various models used to train the dataset and forecast the output in this section.