Despite numerous approaches for code smell detection, the current cutting-edge techniques, such as transfer learning (TL), have not been thoroughly investigated in the domain of code smell detection. This study aims to explore the viability of TL for identifying code smells on homogeneous data by leveraging prior knowledge of previously detected code smells. To achieve this objective, the study has defined the problem of identifying code smells on homogeneous data using TL. The experiment utilized publicly accessible homogeneous datasets that include four types of code smells-long method (LM), data class (DC), god class (GC), and feature envy (FE) to assess the effectiveness of a TL technique called domain invariant transfer kernel learning (DITKL). The evaluation of DITKL's performance has been examined using four performance metrics: precision, recall, accuracy, and the area under the receiver operating characteristic curve. The findings indicated an absence of co-occurrence between GC and DC because of inadequate code smell detection from GC to DC and vice versa. Additionally, the findings indicated the viability of identifying code smells such as LM and FE and revealed a significant correlation between the two. This study suggests the effectiveness of using DITKL to identify instances of code smells and their co-occurrences in experimental datasets when two code smells are related. Therefore, the utilization of TL has the potential to aid in the detection of code smells on homogeneous data in situations where a code smell detection tool or expert knowledge is not accessible.
Software fault prediction (SFP) models are often trained on severely imbalanced data. Prior work also warns that low events‑per‑variable (EPV) ratios can yield unstable and biased estimates. To quantify the separate and joint impact of (i) class‑balancing techniques and (ii) EPV ratios on the accuracy and stability of SFP models trained on five real‑world, open‑source GitHub projects. We also examine whether classifier choice outweighs either factor. We extracted five projects from GitHub and built class‑level datasets. After feature selection (genetic algorithm), we evaluated five estimation techniques (PLS, CART, NB, RF, SVM) under a factorial design crossing EPV ∈ (1) with balancing ∈ (None, Up, Down, SMOTE, ROSE) [1]. AUC is used to measure performance. We report effect sizes and stability (across repeated runs) and profile computational cost. Baselines include (a) no balancing at each EPV level and (b) an additional “raw” baseline without EPV adjustment for the only dataset with non‑zero defects (Ceylon). (i) Balancing improves AUC in a minority of settings; SMOTE is the most reliable of the balancing methods, whereas ROSE is the least effective. (ii) EPV = 20 consistently yields higher accuracy and stability than EPV = 3 or 10. (iii) Classifier choice matters most: strong estimation models (RF/SVM) at low EPV often outperform weak estimation techniques (PLS/CART) at high EPV. For modern GitHub‑based SFP, increasing EPV and choosing an appropriate classifier have a greater impact than applying generic class‑balancing. We provide a reproducible protocol (data processing pipeline, seeds, and scripts) and outline an external validation plan on NASA, PROMISE, and Relink.
The development of autonomous vehicles (AVs) increasingly relies on reinforcement learning (RL) within simulated environments. However, many existing frameworks report only end-of-training metrics, providing limited insight into how design choices influence learning outcomes. In this work, we introduce a unified simulation-based RL framework designed for both parking and driving tasks using the Unity engine. Our main contribution is a custom-built Adaptive Dual-Task Curriculum Scheduler (ADCS), which dynamically adjusts task difficulty to accelerate learning. We also develop a diagnostic logging pipeline that records per-step reward, loss, and entropy metrics, enabling fine-grained performance analysis. By integrating both PPO and DQN agents into this framework, we conduct systematic comparisons against multiple baselines and ablated variants. Our results demonstrate that ADCS, in combination with structured reward shaping, reduces training time by over 50
We present a novel energy-efficient scheme for the wireless transmission of streaming video data from Internet of Video Things (IoVT) visual sensors to the backhaul network. Our solution employs a dynamic reference frame selection mechanism powered by a Long Short-Term Memory (LSTM) deep learning model to implement a low-complexity, lossless video data encoding scheme. By exploiting temporal correlations in video frames of the JPEG and JPEG 2000 standards, our dynamic reference frame selection mechanism creates a lossless encoding of the video data by eliminating redundant information. The encoded data is further compressed using the Redundant Binary Number System (RBNS), resulting in a non-uniform distribution of symbols, with 0’s being the most frequent occurring symbol. A silent-symbol transmission strategy is employed to transmit the resulting RBNS-encoded data, transmitting only the non-zero RBNS symbols (1 and 1̅ ), resulting in significant energy savings. Simulation results on real-world traffic surveillance datasets demonstrate transmitter-side energy savings of over 84% (for outdoor scenarios) and 86% (for indoor scenarios) compared to transmitting raw video files. The proposed method ensures lossless data communication with zero Mean Squared Error (MSE). It outperforms the popular encoding techniques based on neighborhood correlation sequence (NCS) method, discrete wavelet transform (DWT), Haar discrete wavelet transform, high efficiency video coding (HEVC) and AV1 in terms of overall energy efficiency, making it highly suitable for energy-constrained IoVT applications.
A key objective of 6G networks is to bridge the digital divide between rural and urban areas by providing ubiquitous Internet connectivity to underserved rural populations. To achieve this, we propose a hybrid network architecture integrating optical fiber cables (OFC), wireless networks (WN), and free-space optics (FSO), which are cost-effective and easily deployable in rural regions. Our solution addresses critical challenges in rural connectivity, offering a scalable and intelligent framework for future 6G networks. An intelligent multi-access routing algorithm is introduced to dynamically route messages through the most appropriate channel of either OFC or WN or FSO, based on real-time conditions. This approach aims to optimize bandwidth utilization, reduce communication latency, minimize jitter, and reduce packet loss rate. A multi-objective optimization problem is formulated to balance these performance metrics, and its solution is implemented as a channel selection algorithm. The proposed algorithm adopts flow-based routing, where each flow represents a session of multiple packets exchanged between users with specific service requirements. User needs are categorized into five primary service classes, each defined by Quality of Service (QoS) parameters, including bandwidth, latency, jitter, and packet loss rate. A Gated Recurrent Unit (GRU) learning model predicts channel characteristics using historical data, enabling optimal channel assignment to each flow, based on real-time conditions and user QoS requirements. The proposed multi-access hybrid routing scheme ensures efficient allocation of network resources while meeting QoS demands, thereby enhancing user satisfaction.
