Analyzing Consumer Behavior Insights from Smart Meter Data: A Comparative Study of Deep Learning and Traditional Machine Learning Approaches

Therapid deployment of smart meters has transformed the utility sector, offering real-time, highresolution data for improved energy management, grid optimization, and customer engagement. Smart meters, including SMETS1 (Smart Metering Equipment Technical Specifications) and SMETS2 (the second and more advanced generation), record and communicate real-time consumption data, enabling quick detection of faults and irregularities for prompt maintenance. Analyzing the performance and preferences of different smart meter generations, along with non-smart meters, is crucial for optimizing deployment and understanding their impact on energy consumption behavior. This paper aimed to analyze consumer behavior insights from smart meter data through two distinct approaches: deep learning using a Sequential model and traditional machine learning employing Logistic Regression, Decision Tree, and Random Forest algorithms. The dataset, containing information on Grid Supply Point (GSP), Primary location, and various numerical metrics, underwent a thorough preprocessing stage to handle missing values, normalize numerical features, and encode categorical variables. For the deep learning approach, a Sequential model was constructed with input and hidden layers, aiming to predict the number of smart meters (SMETS1 and SMETS2). The model achieved an accuracy of 61% and a loss of 0.02%, showcasing promising results. In the alternative approach, a classification task was formulated to predict whether a location has the highest number of smart meters (SMETS1 and SMETS2) or not. Logistic Regression outperformed Decision Tree and Random Forest models, achieving an accuracy of 98%.