The Effect of Biased Training Data in Self- Driving Cars

Abstract

"Self-driving cars rely on trained machine learning algorithms to navigate the world around them. I present results from experiments that test whether biased training data effects a self- driving car’s ability to identify obstacles in the road. I wrote a machine learning algorithm and created biased and unbiased datasets to train and test the algorithm appropriately. When the algorithm was trained with a biased dataset but tested with an unbiased dataset, the accuracy of the model was poor. This concludes that bias in a self- driving car’s machine learning algorithm does impact the algorithm’s performance and hinders its ability to interpret the world around it."