MACHINE LEARNING ALGORITHMS FOR CLASSIFICATION
In the realm of ML, different algorithms offer distinct approaches to classification, a process crucial for tasks like soil classification or risk assessment in geotechnical engineering.
Consider a music recommendation system where suggestions are based on your past listening habits and those of users with similar tastes. The k-Nearest Neighbors (kNN) algorithm in machine learning functions in a comparable way. For classification in geotechnics, kNN would analyze a soil sample by comparing it with a collection of previously classified samples. The algorithm identifies the ‘nearest’ samples in terms of similar properties, such as mineral composition, density, or moisture content. It then classifies the new sample based on the most frequent classification among these closest samples.
Next, imagine drawing lines in the sand to separate different types of shells. Support Vector Machine (SVM) for classification works similarly by finding the best boundary that divides different classes. For instance, it could be used to distinguish between soil prone to liquefaction and more stable ground.
Decision Trees (DT) in classification categorize data by splitting it based on feature values, like soil composition or density, leading to a final class determination. Each of these ML algorithms provides a unique perspective in analyzing and categorizing geotechnical data, making them essential tools in the field for their ability to discern complex patterns and aid in accurate classification.
SAND CLASSIFICATION
…
PLUGGING OF PILES
In one of our research projects, we are investigating the feasibility of the ML models in accurately forecasting the plugging condition of open-ended pipe piles, a crucial aspect affecting their drivability and capacity. The plugging condition of these piles, which can be plugged, unplugged, or partially plugged, is influenced by their geometrical and geotechnical properties, as well as the dynamics of driving. Traditionally, confidently predicting this plugging condition before driving has not been fully established, often leading to difficulties during pile driving.
To address this challenge, our study focused on using ten different Machine Learning (ML) models to forecast the plugging condition. These models were trained using various combinations of input features, including the piles’ geometrical properties and the geotechnical conditions at the site.
In terms of performance, the traditional method used in the study yielded an accuracy of 53%, while 75% was achieved by the best-performing ML model, showing that the ML models significantly outperformed traditional methods in forecasting plugging conditions. This advancement highlights the potential of ML in providing more accurate and reliable predictions in geotechnical engineering, especially when it comes to complex and variable-dependent scenarios like plugging in pile driving.