From idea to execution: a laptop price prediction model

One of my most rewarding projects during my MSc was laptop price prediction using machine learning. It combines my interest in machine learning with solving a real-world problem, and the full project is open on GitHub.

The problem

Laptops are indispensable, and their prices vary widely by configuration, brand and features. Estimating a fair price is hard for buyers. So I built a model that predicts laptop prices from key specifications such as processor, RAM and GPU.

Overview

The goal was to build regression models that predict price accurately, and compare:

• linear regression with L1/L2 regularisation (Elastic Net)
• Support Vector Regression (SVR)
• Random Forest regression

Random Forest performed best, with an R² of 88.24%, because it captures the non-linear relationships in the data.

How I approached it

Data preparation

The dataset had 1,721 records and 23 attributes: numerical, categorical and binary features.

• Encoding: target, one-hot and ordinal encoding, depending on the feature type.
• Scaling: StandardScaler to normalise the data so the models performed well.
• Dimensionality reduction: Principal Component Analysis (PCA) to reduce dimensions while keeping most of the variance.

Model development

• Elastic Net combined Lasso (L1) and Ridge (L2) regularisation to balance performance and interpretability.
• SVR captured non-linear patterns with an RBF kernel and L2 regularisation for stability.
• Random Forest outperformed both by handling complex relationships and interactions between features.

Evaluation and insights

• Random Forest gave the best result: R² of 88.24%.
• The biggest drivers of price were processor type, RAM, GPU and screen quality.
• Using the trained model, I predicted prices for new, unseen configurations.

Technologies

Python, with Pandas, NumPy, Scikit-learn, Matplotlib and Seaborn, in Jupyter Notebook and Google Colab.

What I learnt

• Data preprocessing (encoding, scaling and dimensionality reduction) does more for accuracy than model choice alone.
• Regularisation is how you balance interpretability against performance.
• Why models like Random Forest suit non-linear relationships, and what you trade for it.

Explore the code

The GitHub repository includes a documented Jupyter Notebook (preprocessing, model building and evaluation), the dataset for reproducibility, and the project report with the full analysis.