Evaluation of Domestic Electricity Cost in Relation to Hydroelectric Energy and Rainfall Using Matrix Completion Method in Kenya

Authors

Samuel Wachira

Correspondent Author

sct2540052021@mmu.ac.ke

Co-Authors:

Clivia Mwangi, Charity Mwaura, Emmaculate Mwangi, Whitney Okore

Abstract

Introduction

Collaborative Filtering (CF) is a machine learning technique traditionally used in recommendation systems to predict the preferences or behaviors of users by analyzing patterns in existing data. It operates by identifying similarities between users or items based on historical interactions—such as rating movies, purchasing products, or selecting content—and making predictions accordingly. CF techniques can be broadly categorized into memory-based and model-based approaches. Memory-based algorithms, like cosine similarity and Pearson correlation, directly compute similarities across the data matrix, while model-based approaches employ machine learning models to capture latent relationships. Though widely recognized for its role in platforms like Netflix or Amazon, Collaborative Filtering has evolved into a powerful tool for uncovering hidden patterns in non-user domains such as energy forecasting, sensor networks, and public policy.

In this research, Collaborative Filtering is applied as a matrix completion technique to predict domestic electricity costs based on rainfall and hydroelectric energy output. By constructing a 3×3 matrix with time-series data covering rainfall (mm), hydroelectric power output (GWh), and electricity cost (KES/kWh) across quarterly intervals, the cosine similarity method (a memory-based CF algorithm), was used to fill in missing or unknown data points. Normalization via z-score allowed for fair comparisons across variables with different units. This enabled researchers to assess the degree of correlation between seasonal rainfall and energy pricing, revealing whether high hydroelectric production, driven by rainfall, translated into reduced electricity costs. The analysis uncovered key insights into the temporal mismatches between rainfall patterns, hydroelectric output, and actual consumer prices, which would not be easily visible through traditional statistical methods.

The integration of Collaborative Filtering in this energy study brings multiple benefits. First, it enhances predictive accuracy in contexts with incomplete or sparse datasets, which is common in developing regions with limited real-time infrastructure. Second, it reveals deeper, non-linear relationships between weather patterns and energy pricing, allowing for better energy policy formulation. Most importantly, the project directly contributes to United Nations Sustainable Development Goal (UNSDG) 7: Ensure access to affordable, reliable, sustainable and modern energy for all. By identifying discrepancies between increased hydroelectric production and persistent high consumer prices, this study provides evidence-based insights for regulators like EPRA to implement fairer tariffs and possibly develop subsidies for low-income households. Ultimately, the approach supports energy affordability and sustainability, two critical targets under SDG 7.

Research Challenges

Publication Details