Hey there! Based on your dataset details (a clear, consistent threshold between male/female weights), let’s walk through the best methods to fill those missing gender values—way better than the linear regression or K-means you tried earlier.

1. Rule-Based Classification (Simplest & Most Effective) #

Since your training data has an absolute, unbroken pattern: all females have weight < 80, all males have weight > 80, this is the no-brainer first approach. You don’t even need a complex ML model here—just a straightforward rule:

• If weight > 80: fill gender with M
• If weight < 80: fill gender with F

How to implement this in your tools: #

• Weka: Use the ReplaceMissingValues filter paired with a custom rule, or use the JRip rule-based classifier (it’ll automatically learn this threshold from your training data).
• Orange: Add a Python Script component and write a few lines of code to apply the rule, or use the Rule Induction widget which will pick up the clear pattern instantly.

2. Supervised Classification Algorithms #

If you want to practice standard ML workflows (great for learning!), this is a classic binary classification problem (predicting one of two discrete labels: M/F). Here are the best fits:

• Decision Trees (e.g., Weka’s J48, Orange’s Tree Widget): Decision trees excel at learning simple threshold-based rules. It’ll split the data right at 80 and give you an interpretable, easy-to-understand model.
• Logistic Regression: Don’t confuse this with linear regression! Logistic regression is designed specifically for binary classification tasks. It’ll model the probability that a weight belongs to M or F, and you can set a threshold (like 0.5) to assign the final label.
• Naive Bayes: Even with just one feature (weight), this probabilistic classifier will work perfectly here—it can easily distinguish the two distinct weight groups.

Why Your Previous Methods Didn’t Work #

Let’s quickly clarify the misfits:

• Linear Regression: This is for predicting continuous numerical values (like predicting weight from height), not discrete categorical labels (M/F). It can’t map a weight value to a gender category.
• K-Means Clustering: This is an unsupervised method—it groups similar data points but has no knowledge of your gender labels. It’ll split your data into two clusters, but it won’t know which cluster corresponds to M or F, so you can’t directly fill the missing labels.

Quick Tip #

Since your training data has zero exceptions to the 80-weight rule, the rule-based approach will give you 100% accuracy with minimal effort. It’s a perfect example of when simpler is better in data analysis!

内容的提问来源于stack exchange，提问作者smcodemeister