Underfitting & Overfitting - Explained In Video



Underfitting & Overfitting

The clip offers a quick look at the main problems in training of machine learning models: underfitting and overfitting. It introduces the aim of machine learning—building a model from data to be able to make correct predictions, and then using this model on new data. A model is underfitting if it is too simple or if it has not been trained properly, which means it cannot find significant patterns in the data and thus will be unable to make good predictions. A more suitable model may be chosen, training may be prolonged, or more training data may be used as solutions. Overfitting occurs in the opposite case, that is where a model is too complex and it “remembers” not only the training data, but also the noise and it mistakes information for which it cannot generalize, so its performance on new data will be poor. The movie also gives hints on how to spot overfitting based on comparing a model’s performance with training data and validation data, hence, reveals significant gaps between training accuracy being high and low validation accuracy.


Comments

Post a Comment

Popular posts from this blog

Application of big data techniques to a problem 4

Image
  Application of Big Data Techniques A high dropout rate — 28 percent of freshmen in one program — is a major obstacle in online STEM education. Teachers collected a huge amount of data to attempt to tackle this, including assignments submitted, posts made in forums, marks on quizzes, views of video lessons and actions in LMS logins. After detecting stress and disengagement indices from chat contributions via natural language processing (NLP), regression models were used to predict who among the students would be at risk of performance dropout. Cohort dropout risk heatmaps, weekly teacher trend reports, and individual student engagement dashboards were applied to visualize this data. These findings enabled more targeted interventions: teachers were automatically alerted to reach out individually to students showing early warning signs, those at risk received specific prompts (e.g., offers of tutoring, inspirational messages, or resources); and teachers were given individual feedbac...
Read more

Application of big data techniques to a problem 1

Image
  Explaining Big Data Concepts Massive amounts of data are coming out of educational platforms because of the rapid growth of online learning, everything from discussion forum activity and video watch time to the frequency of login and assignments submission available. The titanic volume of data, often called “big data,” is being utilized by institutions for the purpose of the improvement of learning results and for the thorough understanding of student behaviour. Edutrack is one such example, where the amount of data annually went up from 1TB to over 10TB, making it necessary to use smarter data strategies instead of manual ones. This comprises, among others, the situation when manual progress tracking was the main method of monitoring, and the end-of-semester survey was the instrument that gave the teachers clues about the best topics for the next semester. The effective scope of the latter was not only limited by its low response rates but also by the time taken to supply an inp...
Read more

Value of Data

Image
Khalid Nawaz 10 March 2025    Value of Data (Including Future Value)  In today’s digital era, big data has become one of the most valuable assets for organizations across industries. It allows businesses to gain deep insights into customer behaviour, preferences, and trends, enabling more informed decision-making. By analysing large volumes of data, companies can optimize marketing strategies, improve product offerings, and enhance customer service. For example, retailers can use purchasing data to forecast demand, personalize recommendations, and manage inventory efficiently. In the healthcare sector, big data supports early disease detection, patient monitoring, and more effective treatment plans by analysing patient records and medical histories.  Governments use it to improve public services, monitor traffic patterns, and respond to emergencies. In agriculture, data from sensors and satellites helps farmers monitor crop health and manage resources more sustainabl...
Read more