# Emphasis on the Minimization of False Negatives or False Positives in Binary Classification

**Authors:** Sanskriti Singh  
**arXiv:** [2204.02526](https://arxiv.org/abs/2204.02526)  
**Date:** 2022-04-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2204.02526v1/

## Key points

- Minimizing false negatives and false positives in binary classification is crucial as machine learning becomes more prevalent
- Existing methods to bias models towards reducing specific cases have shown limited effectiveness
- A new approach involves adjusting the real value of input data post pre-training to improve recall or precision and reduce false negatives or false positives
- The novel method maintains recall improvements while preserving F1 scores, unlike traditional techniques like changing class weights
- Experimentation on multiple model architectures and datasets consistently showed that the new method outperformed traditional approaches
- The innovative approach effectively addresses specific case minimization challenges in binary classification tasks

## Summary

The importance of minimizing false negatives and false positives in binary classification is becoming increasingly crucial as machine learning becomes more prevalent in various products. While existing methods to bias models towards reducing specific cases have shown limited effectiveness, a new approach has been introduced to address this issue without significantly impacting overall performance or the F1 score of the model. This novel method involves adjusting the real value of input data post pre-training, resulting in improved recall or precision and reduced false negatives or false positives across different datasets. To evaluate the efficacy of this method, a simple architecture was employed with a focus on minimizing false negatives. A comparison was made with a commonly used technique of changing class weights, which showed that while class weights decreased false negatives and increased recall, they often led to inconsistent F1 scores. In contrast, the proposed method demonstrated better results by maintaining recall improvements while preserving F1 scores. Through experimentation on multiple model architectures and datasets, it was found that the new method consistently outperformed traditional approaches like class weight adjustments. The results showcased an increase in recall or precision and a reduction in false negatives or false positives without compromising the overall performance of the model. This study highlights the effectiveness of this innovative approach in addressing specific case minimization challenges in binary classification tasks.